Archives

At business meetings over the last year or so, I have a little bet with myself about how long it will take, regardless of the planned agenda, before GenAI is mentioned. Generally it doesn’t take very long. I think that, lately, the record for the longest time before a first mention at work was about 10 minutes. Given that we’re in the software industry I guess that’s not all that surprising. But I got some validation about how widely transformational GenAI technology is becoming at an unexpected place—my recent family reunion.

We have a custom in my immediate family of all getting together for a long weekend in the summer. We gather at holidays too, but in the last 10+ years this summer get-together is something we’ve all come to look forward to. We usually meet, as we did this year, mid-July.

I wasn’t thinking of my workplace “GenAI game” at our family gathering. However, even though only one of our four children is an engineer, it still only took about 2 hours! The first mention came from our oldest daughter, who works in medical marketing in Silicon Valley. She told us how she feeds information and points-of-view to ChatGPT and Gemini, and then challenges each of them to come up with copy addressed to different stakeholders and buyers. She gives pointed feedback to the AI on each iteration as it’s generated, and then has the two LLMs face off with each other, to critique and improve each other’s work. Finally, she manually reworks the best version that the LLMs produce. In other words, she uses GenAI as a sounding board and a talented assistant.

Key to success here is her experience as a mid-career marketing executive. She’s encountered some more junior people who think that GenAI can do it all—and we both agree that someday that might be true. But today, her prior experience doing this “the hard way”—with human-produced messaging, written by herself and by teams under her leadership—is what keeps the quality high. Unchallenged, the output of GenAI is not always that good. We’ve probably all experienced that first-hand reading marketing copy that’s obviously been generated based on minimal information, finding it repetitious and largely content-free. But as a tool in the hands of an experienced person, GenAI is a powerful force multiplier—for marketing as well as for software.

Our next-oldest daughter is a teacher working with high school students (ages 14-18, grades 9-12). She has somewhat the opposite problem: Her youngest students don’t yet have the thinking and organizational skills needed to critically evaluate the content produced by GenAI. In fact, those thinking skills are what she’s trying to teach. When denied access to technology, and forced to use pen and paper, she says it’s clear that some of the 9th graders don’t have the basics of organizing an essay or even a paragraph. With GenAI available to do their homework for them—and some use it even when they are admonished not to—her fear is that these students may never gain the necessary thinking or writing skills. How is it possible to determine what her students actually know, and are able to do, when some or all of their output has been generated by a machine?

I think this is a great question. In my own field, of software development, I think it’s very likely that future generations of developers will not know how to code, in the sense that we code today. They will work at a much higher level of abstraction—perhaps at what we now call the “prompt” level, or perhaps graphically or verbally, to describe the desired behavior to a machine. But based on past trends, I’m very confident that the engineers of the future will still have to know how to think analytically and how to solve problems, even when much of the manual effort that software creation requires today is eventually done by a machine. Future engineers will still require critical thinking skills.

Will future generations still need to write, and coherently organize their thoughts? Or will that human function also be replaced by GenAI?

Personally, I’m convinced that even in a future age of strong AI’s (which is still to come), humans will still need to think critically—and probably to a higher degree than we do today. Even if it’s ‘only’ to judge or guide the output of a machine, the ability to think coherently and to express ourselves succinctly does not seem to me like a skill that humanity can afford to give up. In fact, I think human creativity coupled with machine productivity will lead to greater works in all fields than we’ve seen before: literature, science, art, music, philosophy and, yes, software. Humanity will continue to have a critical role to play in all of these, even in the dawning age of AI.

But the educational conundrum remains: How to teach those critical thinking and expository skills when ignorance can be so easily masked by GenAI? And what about basic factual knowledge—will that still be important? Or can we just look it up or generate it when we need it?

Regardless of the answer to these questions, it’s clear that GenAI tools are here to stay. Education had to find a way to adapt when use of the internet and search engines became widespread—and it did. In my opinion, students now need to learn to use GenAI tools productively—and they also need to learn to think critically and express themselves clearly. The educational industry and I myself don’t yet have an answer on how to do both at the same time, without the one getting in the way of the other. But we’re learning and trying. In a few weeks our daughter is attending a conference on using GenAI in education where they will discuss how to teach both the use of GenAI technology, while also teaching the basic thinking and writing skills needed to use it effectively—and to judge its output. I’m very hopeful.

The youngest of our three daughters is an author and creative writer. In a previous contract assignment, she was also a copywriter for a niche on-line retailer. She used GenAI to first digest large quantities of information on products, competitors, customer feedback, industry trends and other information. She then used GenAI to pull out relevant quotes, and to summarize information from dozens of websites. This sped up her research process enormously (she estimates about 15x), turning it into a simple job rather than a tedious one—and helping her to produce better, more relevant copy faster.

The same daughter is currently doing contract work for an interactive game company in Silicon Valley, writing plots, dialog and characters in genre fiction. She enjoys it a lot, but she and I both believe that at some point this type of more formulaic writing will be possible with GenAI, and humans won’t be needed. Ironically, the human writers employed by the company are currently forbidden from using GenAI—but I think we both see the writing on the wall. In the meantime, she’s enjoying her contract work, and aspires to move up the ladder and become one of the people supervising the writers—whether they are human or AI’s! If that doesn’t work out, she’ll go back to her creative writing. Either way, she sees GenAI impacting her future but doesn’t seem too stressed about it.

Our son, who is an engineer, ironically has the least contact with GenAI. He works for a US defense company, and due to the nature of his work they are actually forbidden to touch it. We had a very stimulating conversation about general artificial intelligence, though, and the gaps in current AI technology that need to be filled to achieve it.

I didn’t expect GenAI to be such a big topic at our family gathering—but it’s clear that it’s beginning to impact all of our lives: for better, for worse—or simply as an agent of change. GenAI is not as pervasive as social media and the internet today, but there’s no question that it is having and will continue to have a profound impact on my whole family, in all our walks of life. We are all in creative professions of one form or another, and we all have ties to tech or to Silicon Valley—so in that sense we’re not completely typical. But I’m sure similar conversations are happening in many families across the world. I think the impact of GenAI on my real-life family is a reflection of the widespread impact this transformational technology is already having on us all.

Postscript: I shared this blog with my family prior to publishing it. Our oldest daughter replied: “Nothing brings a family together like AI!”

At a recent Hitachi Energy conference, I saw a very interesting presentation by Hitachi partner nVidia—the fabless semiconductor company whose GPUs are key drivers of the GenAI revolution. The speaker described nVidia not as a GPU company but rather as a “simulation” company. He described a spectrum of simulation technologies NVidia supports ranging from “physics-based” to “data-based.”

As a person who was educated as a physicist, several light bulbs clicked on for me in this description. What the speaker meant, of course, was that simulations or video games can either be based on ‘algorithms’—that is, a set of physical or un-physical laws (for fantasy worlds, for example)—or they can use extrapolations based on data.

When we as developers write code, we establish a set of ‘laws’ or rules for a computer to follow. Learned behavior, on the other hand, abstracts a set of patterns or probabilities from the data encountered. The latter is the nature of large language models—they are not programmed; rather they are trained based on a selection of natural language text, photographs, music, or other sources of information. 

The models essentially ‘draw their own conclusions’ in a learning process. (Or, more strictly speaking, the models are the artifacts embodying the learning that took place when an algorithm processed the training data.)

Recommended reading: Using AI to Maximize Business Potential: A Guide to Artificial Intelligence for Non-Technical Professionals

Again, this stuck with me very forcefully as an analogy of the human learning process and of the way physics and science work. 

There is a famous anecdote about the physicist Galileo, who was born in the 16th Century, observing the swaying of a chandelier during a church service in the town of Pisa Italy (of leaning tower fame). There was a breeze that occasionally set the chandeliers in motion with larger or smaller oscillations. 

Galileo observed that regardless of how high the chandelier was blown by the wind, once it started to fall, a given chandelier always took the same amount of time to complete an oscillation. In other words, the time the chandelier took to swing back and forth depended only on the length of the chain holding it, not on the height when it was released.

This is quite an extraordinary observation, and the fact that this phenomenon apparently was not noticed (or at least recorded and acted on) for the first 300,000 years or so of human history indicates the degree of insight and curiosity Galileo had. 

Note that Galileo did not have a watch he could use to record the time—they had not been invented yet, and could not have been until this ‘pendulum effect’ had been discovered. Galileo timed those initial oscillations using his pulse—though he later refined his observations using, I presume, the water clocks or sand glasses that were known in his time.

Why is this interesting? Because Galileo, like other discoverers, used observations or ‘data’ to infer patterns. From the data, he was able to make a prediction—namely, that the period of a pendulum depends only on the length of the pendulum, and not on its height of oscillation, or (as was later found) its weight.

Why is this important, and how does it relate to GenAI? There are two broad branches of Physics, called “experimental” and “theoretical”. The goal of experimental physics is to make observations and determine what happens. The goal of theoretical physics is to explain why something happens—specifically, to discover the underlying principles that manifest themselves in observations, or that predict what will be observed.

What is interesting to me in the context of GenAI is that there is a middle ground between these two areas of physics that is sometimes called phenomenology. The term phenomenology is used in different contexts, but back when I was a graduate student in high energy particle physics (theoretical physics, by the way) the word ‘phenomenology’ was used to describe predictions that we did not yet have the theory to explain. 

In other words, we knew that something happened or would happen, but we didn’t yet have a satisfactory explanation for “why.”

Galileo, in his pendulum observations in the church and subsequently in his ‘lab’, was doing what today we would call experimental physics. That is, he was making observations about what happened, and describing what he saw. 

In my limited historical research, I didn’t find a record that he did so, but we can imagine that Galileo could have taken his observations one step further and made quantitative predictions about the behavior of pendulums. That is, based on his experimental results, he could have discovered that for small oscillations, the period of a pendulum was proportional to the square root of the pendulum's length. 

However, even if he had produced such a quantitatively accurate predictive model, history does not record that Galileo ever really understood WHY the pendulum rule he discovered was true. A satisfying qualitative explanation had to wait for roughly 100 years for Dutch scientist Christiaan Huygens’ work on harmonic motion in 1673. A full quantitative explanation required Sir Isaac Newton to first invent calculus and lay out his three laws of motion. (For the theoretical basis of simple harmonic motion, such as a pendulum, see here for example.)

So how does this history relate to GenAI? 

We can readily imagine our current-generation GenAI models acting like Galileo—observing what happens, identifying patterns, and making extrapolations and predictions based on those patterns. We can even imagine them doing the curve fitting and other math required to turn those fresh observations into mathematical models. 

It’s more difficult to imagine a current-generation GenAI model acting like a Huygens or a Newton and inferring from first principles WHY something happens unless the model already contains that information and simply retrieves it. 

I don’t believe reasoning from first principles is impossible for GenAI, and people are working hard on enabling it. Approaches such as “chain of thought” and “train of thought” come close. But ‘theory’ is not the strong suit of current-generation (2024) GenAI technology. Current LLMs are “phenomenologists”, not “theorists”, which is in no way intended to underrate their value.

Why do we care about the theory? If we can predict “what” will happen, do we really care “why”?

This is a good question, and it rapidly gets metaphysical, hinging on the nature of consciousness. Moreover, what constitutes a “satisfying explanation” and “first principles” gets really philosophical fast. But in a practical sense, we can see that both theory and phenomenology have value, each in a different context.

Phenomenology has ‘rough and ready’ practical value. Astronomers and, earlier, astrologers could predict the phase of the moon and the progression of the seasons long before they understood that the Earth orbits the Sun, and the Moon orbits the Earth. These purely phenomenologically-based predictions had a profound impact on human history, including the invention of agriculture which, in turn, led to the creation of cities and civilization. 

But it is the nature of the human mind to try to discern the reasons behind what it observes. People developed theories—initially what we’d now term religious or mythological—to explain why the Sun and Moon behave as they do. They did this many centuries before the discovery of calculus and the law of gravity by Newton; the increasingly refined observations made by Kepler and, earlier, Galileo; and Copernicus’ hypothesis that the earth obits the Sun. It is in the nature of humans to keep asking “why” until a satisfying ‘theory’ is presented to explain the observations.

Watch: Getting GenAI Ready with GlobalLogic

Besides being intellectually satisfying to us humans, the value of theory is that, by reducing observed behavior to an outcome of basic principles, it lets us solve problems and see connections that phenomenology alone does not. 

For example, the theory of simple harmonic motion outlined in the Feynman lecture above not only explains the motion of pendulums (Galileo’s observations), but also the vibration of plucked strings on musical instruments and the movement of weights on springs. When we generalize this slightly, driven harmonic motion (a pendulum pushed by the wind or by the escapement mechanism of a clock) also leads to insights in the area of “resonance”. 

This, in turn, helps us understand diverse phenomena such as the structure of Saturn's rings and the behavior of physical structures like bridges under the influence of an external force, such as the wind. 

By uniting our understanding of multiple observations, a theory helps us discover the underlying connection between phenomena that initially appeared distinct. This process of forming a theory is not confined to physics but is something all of us do in everyday life. We have a theory of the motivations behind our spouse’s or friend’s behavior; as infants, we form the theory that an object continues to exist even when we don’t see it; as students or engineers we form a theory of what it takes to get a good grade or promotion. 

We also form ‘theories’ every day in the software space, when we develop an “architecture” or algorithm that produces a (hopefully) simple system that solves not just one but multiple problems. 

We also abstract out commonalities between diverse systems—for example, logging, observability, and security—and structure them as “cross-cutting concerns” rather than re-inventing them afresh for every system. In general, people consistently synthesize observations and try to discern the underlying cause behind them. It’s our nature.

The human brain functions using a combination of observation, phenomenologically-based prediction, and abstraction or “theory” to understand what it observes and expects. Currently (in 2024), GenAI is strongest in the first two aspects—observation and phenomenologically-based prediction. 

To deliver on the ‘holy’ (or ‘unholy’) grail of artificial general intelligence, AI-based systems need to not only predict but also be able to form abstractions and ‘theories’ based on their observations and predictions. They will need to combine a ‘Galileo brain’ with a ‘Sir Isaac Newton’ brain. 

I expect that we will indeed see such a ‘meeting of minds’ in GenAI, even though we’re not fully there today. We have ourselves as examples that these two modes of thought can co-exist in a single entity. We also know first-hand the power of intelligence that not only predicts “what,” but also understands “why.”

You might also enjoy:

• AI is the Future of Media
• Generations and GenAI
• We’re getting the chance to live in the future

This is probably a well-known fact in sociology or some other such discipline, but it struck me the other day that only the generation that knows how to do something can be the one to make that thing obsolete.

Take driving a car, for example. My generation and the ones preceding me in the U.S. eagerly learned how to drive a car as soon as we were legally allowed. Like most of my contemporaries, I started driver's education as soon as the law allowed, at age 15.5, and had my license in hand as soon as I turned 16 years old. But more recently, the number of Americans receiving their drivers license at age 16 has declined from an already low 46.2% in 1983 to a mere 25.6% in 2018 according to statistic.com [https://www.statista.com/chart/18682/percentage-of-the-us-population-holding-a-drivers-license-by-age-group/]. While not as dramatic a decrease, fewer adults in the US had drivers licenses in 2018 than in previous years.

It’s not inconceivable to me that in a few decades, between the proliferation of ride-sharing services (a technology-driven business model) and  self-driving cars (a new technology), relatively few American adults will know how to drive. This is in the country, the U.S.A., that introduced ‘car culture’ to the world.

But today, in 2024, about 91% of American adults still have a driver’s license. And I think that’s a necessary condition for self-driving cars to evolve.

Any new technology will be imperfect. This means that people who know how to use the previous generation of technology are the ones who need to be the pioneers that introduce the next generation. Those are the people who can revert back to the ‘old’ way when necessary, because the ‘new’ way isn’t quite up to some aspects of the task. While my Tesla does some things very well already, I will still over-ride the self driving features when I believe it’s not doing the right thing. But if I didn’t know how to drive, I would be at the mercy of the car—instead of seeing it as an ally and a tool. Except in controlled and limited (or remotely supervised) conditions, I don’t think a non-driver would feel completely safe in even the best of today’s self-driving cars in all circumstances. But for those of us who can already drive, self-driving functionality is a great thing; we can turn it on or off according to the situation and our needs.

There is no doubt in my mind that self-driving cars will be perfected, and will some day soon drive better and more safely in all circumstances than I do. In some specific areas, my ‘self-driving’ Tesla already does a better job than I would. Today’s children—or (if you’re a pessimist) their children—will truly have no need to learn to drive once they become adults. Except for recreation, I doubt if many will bother learning to drive. Driving ‘manually’ will become a forgotten skill.

But for the present, the only way to perfect self-driving cars is to put them in the hands of people who already know how to drive. Only those people with driving skills can “rescue” the algorithms when they don’t work quite right, or can act as trainers and perfecters of the new technology. In other words, only the people who are masters of the old technology can become the pioneers of the new.

I see the same thing happening in the software industry, as we adopt GenAI-driven development tools. There is no doubt in my mind that at some point in the future, GenAI will produce better code, tests, architectures and other software artifacts than we can create manually—and certainly faster. But as in the car-driving example, only those people with the skills to develop systems manually can be the ones to make the new technology successful.

GenAI-based development will predictably have gaps. While there are some areas where GenAI-driven development can add tremendous value already, it does not seamlessly cover the entire software development lifecycle, and won’t for some time. Humans with ‘traditional’ skillsets are very much required to realize the advantages of GenAI-based development.

For this new technology to succeed, the people who know how to develop software the ‘old fashioned way’ will need to make it successful. But why would we do that? We all believe—reasonably, I think—that this new technology will change our work fundamentally. Why would we risk working ourselves out of a job or, at the least, risk changing our current jobs beyond recognition?

My work experience has shown me, time and time again, that those people who try to make themselves indispensable by withholding knowledge are often the first to lose their jobs in any major transition. We can all probably think of a few examples of people who did manage to avoid being let go in a work transformation by hiding ‘secret knowledge’. But what a miserable existence they must have had! Hoarding knowledge and constantly worrying that someone else would displace them by learning what they think makes them valuable. Such behavior reminds me a bit of Golem stroking the one ring and repeating “My precious!”.

While our generation is the current flag bearer for the accumulated wisdom of software development know-how, its techniques and best practices are far from secret knowledge. Countless books, articles, blogs, training courses, examples and other artifacts exist and can be accessed over the web. When AI’s get smart enough, they have ample material from which to learn—as we’re already seeing. Also, there are fortunes to be made from teaching them, and new job opportunities to be created because of GenAI-native development. There’s no way any of us—or all of us—could hold back this tide, even if we wanted to. GenAI will transform the software industry: that is a given. We can argue about ‘when’ and ‘how’, but I don’t think the ‘what’ is in dispute.

Take heart, though. If you love to drive, I think that even in the upcoming era of truly self-driving cars you will have the opportunity. Manually driven cars will still be available, as an option on new or specialized models, for rental to hobbyists, or through the ‘vintage’ market. Similarly, if you love to program, I’m sure you still can. But our generation will indeed be the generation that makes GenAI-native software development a reality. The only question in my mind is: will it be because of some of us? Or all of us?

I had a stimulating conversation with the head of our GenAI practice, Suhail Khaki, a few weeks ago. Suhail made the remark that the more he works with GenAI, the more it strikes him that it’s less like conventional computer software, and more like a person in the way it interacts. He made the remark: “Intelligence is Intelligence”. That got me thinking: a lot of so-called “issues” with GenAI are actually attributable to the fact that it’s modeled on the way people think. It’s really not GenAI that’s to blame—to a large extent, it’s just surprisingly good at behaving the way we humans do.

If someone asks you the same question on two different occasions, how likely is it that you will give exactly the same answer, word-for-word each time? You won’t, unless it’s some memorized speech. If you ask two different developers to implement the same algorithm, how likely is it that each of them will write exactly the same code? It won’t happen. They may both get it ‘right’, but the two programs will be different—slightly, or even radically.

So why does it surprise and frustrate us when GenAI behaves exactly the same way? Humans give different responses to the same question because many variables influence our behavior, including what we ate for breakfast that morning, who our audience is, how the question was phrased (including intonation), and what we’ve learned and thought about between the first iteration of the question and the second. GenAI has different factors that influence it—no need for it to eat breakfast, yet—but it essentially behaves in a ‘human’ way when it gives a different answer to the same question. Similarly for coding. There are many correct answers to the same software development problem. Which one a given developer picks, or which one the same developer picks on different occasions, are determined by a lot of internal and external variables, not least of which is the sum of our previous experiences and training.

What we call “hallucinations” in GenAI, likewise, are also common to us human beings. In the US, politicians on both sides of the aisle give us ample demonstrations of made-up facts to cover lapses of memory or inconvenient truths. We can argue about whether these political misstatements are deliberate or not, but sometimes human hallucinations are done with no bad intent. An elderly woman I knew had vascular dementia, a brain condition that cuts off access to certain memories or faculties. Her intelligence, however, was largely unaffected. If you asked her about her day, she would happily make up a story about activities that, on the surface, sounded very plausible—but in fact never occurred. There’s no way I believe she did this intentionally, with any attempt to deceive. Instead, absent the actual facts available in her memory, I think her brain creatively generated a response that sounded plausible, but was unfiltered by the truth. She was not diagnosed until she was formally interviewed by a psychologist who asked her objectively factual questions, such as the names of her children. It was only then that it became obvious that she had a medical condition, and that her responses in normal conversation were largely made up.

While I’m not a psychologist, I suspect that human intelligence, when denied access to appropriate information but finding itself in circumstances compelling a real-time response, tends to fill in the blanks—or make stuff up. We’d prefer our politicians and my elderly friend with vascular dementia to simply say “I’m sorry, I don’t know”, “I’d rather not say”, or “I don’t remember”.  But where the person feels an imperative to give an answer regardless of missing or internally suppressed information we get “fake news”, false memories or hallucinations. The same is true of GenAI—it defaults to a plausible-sounding but invalid response when it can’t find an accurate one.

My wife is a psychologist, and she tells me that in the human brain there is concept called “filling in the missing gestalt”. The brain tries different strategies and options to fill in missing data. This presentation of options contributes to human creativity and problem-solving behavior. We’ve all experienced this when we’ve been puzzled trying to solve a problem, and then suddenly the answer comes to us. This happens largely sub-consciously, below our level of awareness. When there is insufficient rejection by our brain of wrong alternatives, then we can get human confabulation to “fill in the blanks”, even though the best option might be to ‘leave it blank’, and say you don’t know. But where our brains make a good choice among the generated alternatives, we get originality, spontaneity and invention.

In an LLM, this is controllable to some degree by setting a parameter called the “temperature” which, essentially, governs the degree of randomness used to generate alternative responses. While lowering the temperature limits hallucinations in an LLM, it also reduces the number of good alternatives that are being considered. The downside of fewer alternatives is that the ‘better’ and ‘best’ alternatives may not be generated at all, limiting the effective ‘intelligence’ of the AI. Rather than surpressing the generation of alternatives, the right answer, in my view, is better filtration of multiple generated alternatives. Indeed, a number of GenAI startups are working on hallucination prevention by intelligently filtering generated responses. But the generation of alternative responses, even wrong ones, is actually a characteristic of human-type intelligence itself—it’s a “feature”, not a “bug”. We’re just at a relatively early state-of-the-art in terms of filtration—though I’m convinced that is coming.

Why do these ‘human’ inconsistencies and confabulations surprise and annoy us when it comes from GenAI? Most of us have grown up with computers. While they can be frustrating or bewildering to deal with at times, computers are also predictable. In particular, when programmed, computers do the same thing in the same way every time, and consistently give you the same answer to the same question. We experience computers as machines or ‘robotic’ (in the narrow sense) in the interactions we have with them.

GenAI is not that way. While it runs on a machine, it acts in important ways more like a person. Compared to a programmatic device, GenAI is relatively unpredictable and inconsistent.

I would argue that the unpredictability and inconsistency of GenAI is an essential feature of any intelligence that tries to emulate, in some respects, the human brain. Perhaps inconsistency is a feature of intelligence in general. It may not be a feature we always like, but if we want the advantages of intelligence in our machines, I think we will also learn to make do with its quirks.

Does that mean we can’t use GenAI for useful work? I would argue that, despite our own foibles and lapses, we have used fallible people to do useful work for many, many generations. We can follow some of these same practices in using GenAI.

When managing people, we often have multiple specialists assigned to different aspects of the same activity. Often work is overseen by a manager, who ensures the consistency and quality of the output. For critical tasks, we have documented procedures that people are required to follow. And in emergency situations, or those requiring real-time body control (like sports), we fall back on training. Trained responses are those where people follow pre-defined or pre-learned guidelines—essentially programming—automatically, and largely without thinking. These same principles of human work can, and are, being applied to GenAI today.

Consciously or sub-consciously, analogs to human organization are being developed and applied to GenAI’s today, with more in the works. “Ensembles” of specialized LLMs are being orchestrated by “agents” and other technologies to leverage the strengths of each model, analogous to a human team with complementary skillsets. Like a human supervisor, GenAI management approaches such as “LLMs for LLMs” and programmatic analysis of model outputs, are emerging to filter and  evaluate the quality of an AI’s output. These managers can also trap hallucinations, and send the AI’s—or team of AI’s—back to the drawing board to come up with a better answer. For critical or end-user facing tasks, implementations may combine the best features of programmed approaches and GenAI models. For example, a customer support application might use dialogflow [https://cloud.google.com/dialogflow] for the structured element of dialogs, together with one or more LLMs for call routing, information gathering and results summarization.

The final frontier is, perhaps, machine or industrial control systems, or control of life-critical real-time systems. For these systems, we need deterministic outputs. Creativity may be useful in some situations, but even with humans in the loop, we generally used trained responses and documented step-by-step procedures that we expect people to robotically follow. This is because in an emergency there is rarely time or mental energy to improvise—and documented procedures have been researched, vetted and tested. The robotic following of directions is probably the least human thing that we do, but it’s necessary sometimes—for example, in an emergency situation like steering your car out of a skid when it slips on the ice. Improvising from scratch is the wrong approach in that case—we’re better off if we have trained ourselves to turn into the skid to regain control, without having to process the physics in real-time. For activities like sports, piloting an aircraft in an emergency, and others real-time decision-making, learned and trained skills are an important foundation. Creativity is still beneficial at times, but only on top of a firm foundation of learned skills.

Like human trained responses to emergency or real-time sports situations, control systems operated by computer tend to be more automatic, rules-based and deterministic. That’s not to rule out AI entirely. We already have good examples of conventional, non-GenAI models playing an important role in such systems: For example, your car’s advanced driver assistance controls for lane-following, collision avoidance and adaptive cruise control have important aspects that are AI based. My experience is that these AI-based systems add substantially to my safety. However, I think all of us, myself included, would hesitate before putting our lives in the hands of a technology that is subject to hallucinations. On the other hand, I drove myself for many years without any driver assistance at all, and my fallible human brain still enabled me to survive. So maybe—properly supervised—GenAI has a role here too. After all—intelligence is intelligence, even if it’s artificial.

I was one of the early buyers of the first release of Apple Vision Pro AR headset early this year. I got up at 5am my time to place an order on-line at the first moment when the device became available for pre-order. I then made an appointment at my local brick-and-mortar Apple Store to pick it up as early as possible on the first Saturday after they shipped. Needless to say, I was excited about this technology (and still am).

I got to the store early and waited in line with other eager Vision Pro buyers also picking up their orders. When the store opened and we all filed in, we were each assigned to a store associate for a demo and familiarization session with this new device. While I was overwhelmed by the experience offered by the device itself, what struck me just as forcefully was the nature of the retail experience. The air of excitement in the store was palpable, not only among the group of buyers, but also among the sales associates and even the store manager. I felt like we were all conspiring to share something truly unique, truly special. And that, I think, is the essence of a great retail experience: It’s a conspiracy between the buyer and seller to share a thing of value.

I’ve had this experience a few other times in my life, generally with the owner of a “Mom & Pop” or family enterprise. I particularly remember an experience in India where I bought a leather wing-back chair. I had seen the chair on display at the old Bangalore airport, and because I couldn’t find a retail outlet, I made a visit to the local factory where it was made. The owner of the factory personally showed me around, and actually took my measurements, like I was buying a suit of clothes. They made the chair “to order” and to fit my dimensions (hip-to-knee, length of torso, etc.)! When his workers brought the finished chair to the company apartment where I was staying at the time, you could feel their pride in the finished product. They waited eagerly until I sat in the chair and pronounced it a perfect fit. Years ago, I brought the chair back to the US with me, and it sits proudly in my study to this day, full of very pleasant memories around how it came into my life.

Why are retail experiences like this rare?

Well, when I’ve had them, I was buying something that was at least relatively expensive, and from someone to whom the sale would be significant in some way. In the Apple Vision Pro example, I assume that—in addition to the sales people’s genuine excitement around this new product—the store probably had incentives, a contest, or at least recognition in place for their associates to first learn and then present this new technology. I don’t know that for a fact, but I did get a survey soon afterward, from which I assume (hope) the sales associate was recognized in some way for his great work. In the factory owner example, in addition to their genuine pride in a great product, I think they were hopeful that I’d spread word-of-mouth among my colleagues and fellow ‘ex-pats’ (which I did). In other words, even beyond the money they received, it was worthwhile for someone to deliver an exceptional product and experience.

Notably, none of my best retail experiences were entirely on-line. This being the 21st Century, they all involved the Web: I originally ordered the Apple Vision Pro on-line, and I tracked down the manufacturer of the chair through Google. Other “peak” retail experiences have all likewise involved the Web in some way—education, awareness and so on. But none of my best retail experiences to date have been entirely on-line; all have required a personal touch.

Why is this? Thirty years after the Web came to prominence, why is it not delivering such ‘peak’ retail experiences by itself?

I think it is because of the relative lack of personalization. A critical factor in any ‘conspiracy’—which I believe a great retail experience requires—is collusion between one or more people. Sophisticated websites use machine learning and demographic data to present an experience that is in some ways tailored to the buyer—or at least to a similar type of buyer. However, in my own on-line shopping, I have not yet encountered a retail website that really directly engages me, individually, the way a good human retailer does. My best experiences to date have all required a “human in the loop”.

Back in the early days of the Web I was inspired by a then-recently-published 1993 book called “The One to One Future: Building Relationships One Customer at a Time”, by Don Peppers and Martha Rogers. Peppers’ and Rogers’ thesis was that businesses can prosper by building strong relationships with their best customers, and by growing their share of that customer’s wallet over time. When my team was developing an early e-commerce platform at Apple in the 1990’s, we built in a number of features that enabled automated individualized shopping experiences—such as intelligent cross-selling. But, while we aimed to achieve it, the technology did not exist at that time to truly enable the kind of “mass personalization” and personalized relationships that Peppers and Rogers envisioned. To this day, thirty years later, developing a deep personal relationship with a customer has required a person—a human in the loop. It’s not something the industry has truly been successful at automating.

Earlier attempts at automating ‘personalized’ customer interaction, such as chat-bots and IVR systems, have had limited success. This is largely, I believe, due to the necessarily scripted and programmed nature of these interactions. While some experiences are far better than others, people do not interact naturally on a ‘programmed’ basis—it tends to sound phony unless the system vocabulary is very large, and the options are truly flexible, or even dynamic. That’s exactly what we have in GenAI—interactions that are increasingly open-ended and therefore more human-like.

Could a GenAI-based retail system learn to understand me as an individual—what drives me, what decision criteria I use, what I value in a product? And then map that information to identify the products I would love to buy, and help me understand why I’d love and can afford those products? I believe the answer is an emphatic “yes”, going far beyond traditional ‘propensity to buy’ determinations. The main wildcard here is access to accurate customer behavioral and motivational data and, ultimately, the customer’s willingness to be known deeply by a given system. In general, people are more disclosing to a machine than to other people, because the fear of judgement is reduced. People’s main concern when disclosing information to a machine tends to be the privacy of the information, and concerns about the manner in which that information will be used. These are solvable problems. In particular, I think shopping ‘agents’ will be developed who understand a customer and a customer’s finances and goals intimately—perhaps presenting an anonymized face to the world, for privacy sake.

The other wildcard is how disclosing a retailer is willing to be about the true nature of their products. Mapping an individual customer to a product description only creates loyalty if that customer is truly delighted after receiving and using the product itself. Marketing hype may help sell a product once. But alone, it doesn’t create the kind of “lifetime customer” or increasing “wallet share” that Pepper and Rogers describe in the “One to One Future”. Thankfully, we live in an age of increasing transparency, with customer ratings and reviews now a routine part of the buying experience, and suppliers increasingly honest about sourcing, provenance and other criteria of concern to various individuals. While these systems and data can be—and often are—gamed, they point the way toward a more objective and satisfying retail experience in the future.

I remember Steve Jobs once sent a thoughtful email to the employees at NeXT talking about how retail experiences always involve individual ‘values’—not in a moral sense necessarily, but in terms of the importance they give to certain factors. He gave the example of himself and his wife Laurene considering the purchase of a high-quality European washing machine for their home. He said that the washer produced softer clothes and used less water and laundry soap, and was therefore better for the environment. On the other hand, the washing cycle was longer, so they couldn’t clean as many clothes in the same amount of time. Steve and Laurene opted for the European machine I believe. But you can readily imagine someone with a large family valuing the higher throughput of a cheaper US-made machine above the quality and environmental advantages posed by the more expensive European option. With its natural language understanding, consideration of these kind of value judgements are a very good use case for GenAI. It’s exciting to think of the opportunities for delightful retail experiences for us all, both as consumers, and as sellers, in the one-to-one GenAI future.

Also with GenAI, we have an expanded opportunity to not only create personalized experiences, but personalized products as well. As a simple example, on request a telco analyzes a specific customer’s historic data and voice usage, media viewing habits and other factors—such as payment history—and dynamically creates and prices a product tailored to that specific individual. On a slightly more speculative level—but by no means far-fetched—physical goods can be produced to individual order using GenAI-produced programs and models to drive a combination of industrial robots, 3D printing, CNC milling machines and other computer-controlled manufacturing devices. These technologies exist today and, indeed, are being used to produce products under GenAI control in isolated pockets and on a small scale already. The only speculation is that these approaches will become affordable and widespread.

While it has taken thirty years and counting, I think the technical basis to realize a true “One to One” future is now in place. I think the next thirty years will transform retail almost beyond recognition, by allowing people and AI’s to conspire together to create value and deliver delightful buying experiences. And, not incidentally, create tremendous customer loyalty to those who deliver this service.