How could something so perverse come to pass in what used to be a science-based industry full of level-headed executives you ask? Well it couldn’t happen had that industry not abandoned science and jumped head first into a bizarre new hype-driven mythology - where untruth and fiction rate higher than architecture and intelligence. Today’s alleged great innovators want to believe, not to create and Wall Street is rewarding their every sci-fi meme-based announcement as if it were gospel. And driving all of this are the myths surrounding Artificial General Intelligence (AGI) - the more or less agreed upon end goal of all the wanna-be trillionaires.

The Ten AGI Myths Driving The AI Industry

AGI Myth #1 - AGI (at any level or in any form), will be achieved using a single (AI) model architecture. And the majority of tech leaders who are saying that claim this will be be done through LLMs; although the few positing World Models, Reasoning Models or even RL aren’t doing much better in that they mostly still insist it’s just one type of model that will get us to AGI.

Why won’t this work? Because this supposition ignores the most basic metaphors associated with human intelligence, which ought to be our guide; a) that our mental 'architecture' is in fact a composite of various components - even if those components utilize similar building blocks and b) human intelligence does not exist in a vacuum - it's at least partially 'collective.' An overly simplistic route to AGI may seem easier to manage, but ultimately will fail to reach its objective.

AGI Myth #2: That Artificial General Intelligence (AGI) can be achieved without an industry-agreed upon definition of what it actually represents. Such a definition is neither self-evident nor irrelevant - it is in fact critical to every aspect of both design and evaluation.

Without one, we will continue to be plagued with non-stop hype and false claims. This missing definition joins a long line of poorly or wholly undefined terms in the AI Industry; including - Intelligence, Agent, AI, Learning, Thinking (in the context of AI), etc. The lack of such definitions is why we tend to get wild and unsubstantiated claims like LLMs can learn and determine right from wrong.

AGI Myth #3: That Artificial General Intelligence (AGI) is or more accurately will become one thing. There will instead be many AGIs (some competing, others cooperative) and they'll exist across a spectrum of capability (still general in nature) and will also coexist alongside a growing ecosystem of targeted intelligences. All of this presupposes agreed upon definitions of course.

AGI Myth # 4: Human Intelligence is not General, and therefore may not be the appropriate metaphor for AGI design. This is an interesting one and reflects the fact that not only can't we define AI or AGI properly, we're also still struggling with Intelligence as a core concept.

Human Intelligence is in fact the closest we may ever get to a true General Intelligence because a) any such intelligence requires a central identity (or self) to function, b) human intelligence can solve problems on both the individual and collective level (which often is forgotten when having this argument), c) Human intelligence leverages abstract thinking in combination with real-world interaction (which is precisely what is needed for learning).

Yes, it’s possible perhaps to abstract a more generic general intelligence away from the human metaphor, but why exactly would we want to do that when in fact we’re trying to make these solutions; a) more predictable and b) easier to work with?

AGI Myth #5 - Even though Human-like Intelligence may be the right design metaphor for reaching AGI (see myth # 4), that doesn't mean that humanoid robots are required to make it happen. While the two might be combined successfully someday, AGI will follow its own path towards actualization.

This is where myth intersects with the many sci-fi memes being borrowed by today’s class of wanna-be trillionaires. Humanoid robots (even though they’ve never actually existed outside of sci-fi) seem at first glance to be the easiest way to interact with an AGI and assure that the intelligence has proper sensory inputs (at least that seems to be the premise). In the real world, they’re incredibly hard to create and ultimately aren’t necessary to gain the benefits of a general intelligence. Worse than that though is the question as to whether they’d end up being incredibly self-defeating economically. In other words, there is neither a technical mandate nor true economic incentive to follow this path.

AGI Myth #6 - Hyper-scaling LLMs will automatically, magically bring about AGI (and how it accomplishes this magic isn’t really relevant). This is perhaps analogous to the notion that strapping a jet pack on a Model T will produce a flying car (Tesla's next product, driven by robots? who knows).

The Architecture of AI follows the same rules as every other IT Architecture (there is nothing magical about any of it) - thus, AGI must be "Intentionally Designed" to fulfill the expected capabilities (even if that ability is self-learning general intelligence). Hyper-scaling is entirely based upon a misunderstanding of the existing limitations of LLMs and one or two papers. It will turn out to be the most expensive false assumption in history.

AGI Myth #7 - AGI is the path to Super Intelligence. It's problematic enough that we don't yet have adequate definitions for either of these, but the larger paradoxical question here is why producing AGI (or human-like) intelligence would automatically lead to super human intelligence even though that didn’t happen with natural intelligence - it's a counter-intuitive conclusion at a fundamental level. And going further, has adding human collective knowledge together yet led to this for humans (that would be our version of hyper-scaling)? Nope. Super intelligence is something else, what that is, we don’t know yet.

Why don't industry leaders point out the more obvious implication; that AGI will lead to human-like artificial entities? Now, this is fairly profound in itself, but it's not super intelligence or dystopia. If those entities become Nazis after reading Grokipedia that’s potentially a problem, but not a civilization ending one.

AGI Myth #8 - AGI will inevitably lead to the "Singularity." This is one of the dumbest myths, (right up there with hyper-scaling LLMs gets you to AGI), and involves not just a an automatic jump to super-intelligence, but basically goes straight from AGI to synthetic God. And of course, does anyone really know what Singularity actually means in this context (of course not). They’ve borrowed a term from Physics to make themselves sound cooler and smarter than they actually are.

The countless Science Fiction tropes that immediately jump from AGI to an instant AI apocalypse may be entertaining, but they never did have realistic science behind them and they still don't. But that hasn’t stopped multiple AI company CEOs from proclaiming that it’s right around the corner which of course distracts from all of the real-life nightmare problems that they don’t want to deal with now.

AGI Myth # 9 - That reaching Artificial General Intelligence is inevitable. While it is likely, this 'inevitable' achievement presupposes that innovation, science and common sense will prevail over the mindless hype now rampant in the AI industry. There is also a growing possibility that they won't.

The impact of not reaching AGI after all of this epic hype will be disastrous for the AI industry and the whole economy (not because other forms of AI won't be valuable, they will) but due to the massive mismatch of expectations and financial collapse of the industry.

AGI Myth #10 - That the current tech company leaders in the AI industry are capable of bringing AGI about. While there are some actual thought leaders in the industry who are headed in the right direction, they're all on the outside now of the hype / financialization axis that is driving every major AI and Tech company involved in the pursuit of AGI (in the US).

Other countries aren't necessarily burdened with this level of misdirected groupthink however, and the longer that the US remains stuck where we are, the more likely someone outside the US will be the one making the necessary breakthroughs. To put it in the Administration's terms - the US is now losing the AI / AGI race. We are winning the race to build the greatest number of useless data centers in the fastest time, however.

Conclusion

We’ve now entered an age of myth and unreality (and not just in relation to AI technology, it’s carried over into and intersected with the growing political unreality as the tech bro leaders also become political movers and shakers). In this fantasy land of sci-fi supposedly come true, an Elon Musk can announce (as he did last week) that launching 1,000,000 Starlink satellites would provide a cheaper solution for solar power (for orbiting data centers) than building solar farms on earth for earth bound data centers (having claimed that the earth was already too saturated to build on anymore). Instead of universal mockery, that proclamation (along with the announcement of data centers on the moon) was taken seriously by most of the world’s media and even investment groups like Morgan Stanley. Something profound has changed in our society for that to have been the reaction - the tech gods are now orchestrating a new reality from atop Mount Olympus.

Copyright 2026, Digital Perspectives

AI’s Successes (to date)

• AI has made significant breakthroughs over a twenty year period, but not necessarily revolutionary ones (at least in regards to achieving actual Intelligence).

• The closest to a ‘revolutionary’ breakthrough has come through Generative AI - which has succeeded in mimicking (and indeed surpassing) some aspects of creativity which we had always believed would largely remain the domain of humans. It is not itself creative in our sense though, as it has essentially regurgitated the work of humans as opposed to ‘imagining’ something on its own.

• AI is providing additional augmentative and automation support for a variety of jobs / tasks which will prove highly consequential over time (even if it has not fully sunk in yet or been appropriately supported through reasonable Use Cases).

AI’s Failures (to date)

• AI, even where it has been successful - has a hit a wall. Scaling will no longer improve it much, more data will no longer improve it much (because much of that will be fake from now on - they call that “Synthetic Data”) and it just still has problems doing a lot of basic things properly. All of the big AI companies know that they’ve already hit this wall and they’re absolutely, positively terrified about it. This is precisely why they keep trying to change the dialog away from what’s happening with the models themselves, (by obsessing over hyper-scaling or predicting doomsday etc.).

• Artificial Intelligence has not achieved any level of human-like / actual intelligence. Not only that, the AI industry steadfastly refuses to help define what that is or would be if translated to a machine (which in turn facilitates their ability to claim intelligence w/o s real way of actually confirming it).

• The current GenAI-based successes will not translate into superintelligence or AGI. We’ve talked about this at great length here on Digital Perspectives, but the easiest way to understand why this is case is that the current success has been somewhat accidental and that a more profound level of (artificial) intelligence requires two things before it can really happen; 1) that missing definition of what it is and 2) a deliberate design to get us there - thus we have to have both and the second is dependent on the first. Now “design” in this context does not mean that all of the logic is coded (as in the old days), but rather that there needs to be an “Architecture for Intelligence.”

AI as Finance rather than Innovation

In the midst of all of these other issues (and partially because of some of them), the primary focus for the entire AI industry has shifted over the past 3 years from solving AI innovation hurdles to inflating AI company valuations up to astronomical proportions. We’ve talked about this here too - the onset of AI Groupthink and the death of scientific or rational thinking in relation to AI industry (and now national) policy and objectives are leading us down a very bleak path. Yes, it’s important that technology makes money for those who invest in it - but of course the AI itself isn’t making much money (and there doesn’t seem to be any plausible revenue path for making up all of the more than 1 trillion being invested now short of perhaps replacing every worker on earth) - all of the real money connected to AI is tied up in Hyper-Speculation.

Hyper-Speculation - This is a relatively new phenomenon (which started in the late 1990’s with the Dotcom & Telecom bubble) where the Price to Earnings (P/E) ratios for Tech companies shot up to huge and unsustainable levels based on massive hype coupled with unrealistic expectations for the technologies in question. This led to the profit opportunities shifting from the actual products themselves to the financialization of them. In other words, Tech stocks become ‘investment schemes’ that are more or less disassociated from whatever the company might do in the real economy. While at the underlying core of this financialization, there is a real company - so it’s not completely made up like a Ponzi Scheme - it essentially behaves like one anyway.

Yes, all companies (or most companies) get securitized and enter financial markets - but when the P/E ratio becomes detached from the company fundamentals (and economic reality), it distorts perspectives (both from inside the company itself and also that of everyone viewing it from the outside). That distortion drives decision-making and by and large (historically speaking), bad decision-making. If for example, OpenAI (which has a current valuation of $500 billion on profits of maybe $13 billion) were openly traded, it’s P/E ratio would be around 38, and some other Tech companies are in the 100+ range (which is about 20x higher than other type of company). The decisions made in these types of scenarios (where the P/E ratio is reaching for the stratosphere) are then based not on what gets the company to a profitable product, but what gets the valuation even higher (because of all the money that is being made on trading and funding deals as opposed to on revenue, which actually might not even be necessary). Many companies during the Dotcom bubble made 100’s of millions for their founders and then simply vanished (much like Elon Musk’s first company Zip2, which got him a $22 million payout for vaporware - we’ll be looking more closely at how Musk has mastered the Financialization Scam that the AI industry has adopted in an upcoming article).

How is this Wrecking The AI Industry?

Well, there are several ways that this immediate wealth gratification response by AI company owners and CEOs is destroying the long-term potential of the AI industry:

• The Mismatch of Expectations: There is no faster way to turn off customers than to promise something that a product simply cannot do. AI companies claiming that LLMs can be AGIs is a little like saying that your vacuum cleaner can build a skyscraper; it’s simply a ridiculous boast that’s not attached in any way to reality. That’s not to say that current AI can’t do anything - it surely can. But the gulf between the AI hype and the actual value proposition is growing ever wider and it’s already catching up to the big AI players. Now, in the current environment where financialization rather than revenue matters - these companies probably don’t care - but they should because this phase can last only so long before it implodes.

• Going down the Wrong Path: And if those companies (and the country), actually do wish to someday achieve machine intelligences that comes closer to AGI etc., then the choices that they’re making now will make that harder as $ trillions, yes trillions of dollars are sunk into flawed architectures which will be either burned out or obsolete in a year or two (there is no long-term infrastructure dividend here as there was in the Telecom bubble - AI chips and mega data centers are perishable commodities).

• Wrecking the Real World: The other side of the hype problem is that many people (far too many), still believe it. Many of them run companies and they think that this technology is ready now to replace a huge chunk of their workforces; the problem of course is that it isn’t. But the AI not being ready yet isn’t stopping these companies from firing 100’s of thousands of people right now. Unlike automation in manufacturing, (where clear evidence drove labor replacement decisions), in this case businesses are interpreting the hype cycle itself as the necessary evidence. This process in turn is setting off a massive negative feedback loop that’s accomplishing two very destructive things simultaneously; 1) it’s handicapping those companies in ways that they don’t fully comprehend yet (not unlike the way Elon handicapped the US Government last year by arbitrarily firing 329,000 people through DOGE), and 2) the cumulative effect of 100’s of thousands or millions of people getting fired is already hitting the economy - hard. Both of these outcomes will hurt those companies and then that will bounce back onto the AI and software companies dependent upon them for business Thus the hype feeding the financialization wealth machine eventually kills the real one.

There is no outcome associated with the current path of the AI Industry that leads to success (for them or for us). Every hyper-scaled data center that these companies build adds a burden to the communities that they’re infesting w/o any new jobs added; they will also fail to make the AI models perform any better. And every dollar invested (and lost) going down the wrong path (full of sunk costs) makes it increasingly unlikely that there will be money or motivation left to go down the right one. Those AI company CEOs driving their not-so-intelligent cars over the cliff at full speed may not really care about any of this if they manage to get their $100’s of billions out before their companies (and the global economy) collapses - but the rest of us definitely should care.

Copyright 2026, Digital Perspectives

Moonlight Graham can’t hang up his stethoscope just yet, the Field of AI Dreams is costing a fortune and threatening the farm with bankruptcy (think of our economy as the farm), and it’s not too clear whether any of the spirit players are even showing up yet (well, maybe we’ve got Joe Jackson so far, but not the rest of them). The metaphor of Moonlight Graham is actually more pertinent than you might think given that OpenAI just launched a medical product, “OpenAI for Healthcare.” Is it ready for primetime or “The Pitt,” who knows - when you go to their site there’s almost no detail about what the service or product actually does. This is problematic on many levels, especially when you look and see that the main thing being offered is a ChatGPT for Healthcare. Obviously there’s a lot of potential there too, but the evidence and practice support for this thusfar seems pretty thin to be rolling it out in the manner that’s currently being done. And this is all part and parcel of Big Tech’s seemingly desperate need to prove AI’s immediate value and actually generate revenue (something that’s not been happening at anywhere near the pace of AI investment).

And even more preposterous claims and plans were being hatched over the last several weeks as well; namely the data centers in Space lunacy. Elon Musk is trying to push his net worth over a trillion $, so he’s got make up something new these days. He tells us that well, we won’t have to worry about power because of - the sun! Wow. Never mind that the International Space Station cost $150 billion and $3 billion a year to run and is the size a big house and that we’d need something about 1,000 times larger in orbit - but hey - free solar power! There is level of insanity that’s crept into Big Tech that’s simply never existed before and it’s all revolving around AI.

Hilarious AGI Definition & Disclaimer from Sequoia Capital - “Blissfully Unencumbered by the Details. Before we go any further, it’s worth acknowledging that we do not have the moral authority to propose a technical definition of AGI. We are investors. We study markets, founders, and the collision thereof: businesses Given that, ours is a functional definition, not a technical definition. New technical capabilities beg the Don Valentine question: so what? A Functional Definition of AI: The answer resides in real world impact. AGI is the ability to figure things out. That’s it.*

No, that’s not it and No We Don’t Have AGI or Superintelligence, Yet.

Saying that we do have it and then disclaiming, “hey don’t worry, be happy and get rich and it just don’t matter what it is anyway because we’re just investors,” seems to be about the most disingenuous message crafted in a Valley full of hype and half promises ever.

Here are the Top 5 reasons why we don’t have either (AGI or Superintelligence yet):

1. Contrary to the dismissive rationale above, definitions do matter. You cannot build something to a specification that does not exist. “Oh, but our stuff works so great that it is has to be AI, has to be AGI, has to be Super-intelligent, right …” well so did every other IT tool for the past 80 years - so what, not only that, we knew what we were building to then and how it worked.

2. You can’t properly evaluate something that no one agrees upon.

3. The folks making $50 to 100 trillion on something may not be the impartial judges needed to perform such evaluations even if we did have an agreed upon approach, a) with a definition and b) a tailored evaluation paradigm based upon that definition.

4. Because the immediate experience of most people AI using these tools now (and we’re talking about now, not 5 years from now), seems to contradict these claims. The performance of LLMs (even the newest ones) are failing at basic AGI-related assessments while the companies are focusing more on the benchmarks that they can cherry-pick. So, while we may have some level of Artificial Intelligence at work (and even here the definitions aren’t that great), it’s nowhere near being “General” - in fact, most of these tools are far more specialized than most people realize.

5. Because the credibility of the industry is now suspect in relation to their expectations as to what type of solution will bring about AGI. In other words, if it requires data centers in space, and data centers the size of Manhattan and 1/3 of the entire US power grid to run this stuff, maybe there’s a pretty big flaw in the current design approach - a flaw that not only makes the current AI the most inefficient IT solution ever, but one that’s not well suited to ever be a “General” intelligence.

In The Field of Dreams, Kevin Costner and his family succeeded because they were banking on magic and mystery and for them it paid off - but Silicon Valley isn’t the right place for everyone to begin depending upon magic happening. If AGI remains the mystery that Sequoia believes it to be, the only thing that will save them and their money may be a miracle.

Copyright 2026, Digital Perspectives

Some of the previous articles where we’ve talked about this topic include:

• What is (AI) Reasoning? part 5 - Architecting by Intention, not Accident.

• What is the AI Revolution?

• A Framework for Evolutionary Artificial Thought.

• Orders of Artificial General Intelligence (AGI)

Definitions

We’ll go over some of these definitions again to establish the frame of reference…

The Turing Test, originally called the “Imitation Game” by Alan Turing in 1949, is a test of a machine’s ability to exhibit intelligent behavior equivalent to that of a human. In the test, a human evaluator judges a text transcript of a natural-language conversation between a human and a machine. The evaluator tries to identify the machine, and the machine passes if the evaluator cannot reliably tell them apart. The results would not depend on the machine’s ability to answer questions correctly, only on how closely its answers resembled those of a human. - Wikipedia

While, the Turing Test isn’t actually how most folks now measure Artificial Intelligence, it still sits at the core of many (if not most) of the associated expectations for AI Evaluation. Some of those expectations include:

1. The ability to mimic human interaction, behavior or responses.

2. Use of variants of IQ testing (which these days have become complex problem-solving measures, but still harken to the earlier paradigm).

3. The Self-Fulfilling Prophecy Effect; this involves the assumptions that in order to mimic human behavior or responses, an ever greater pool of examples are required (for prediction) and thus an ever-increasing model is required to process that massive pool of examples and it must be housed within ever-increasingly huge data centers. All of this does actually trace back to the expectation that an AI model’s output needs to ‘sound like’ a person to prove intelligence.

AI Evaluation, or "evals," is the crucial process of systematically testing and measuring AI system performance, accuracy, and reliability against specific goals, using structured tests and metrics to ensure they function as intended, avoid issues like hallucinations, and deliver value, a necessity due to AI's variability compared to traditional software. Evaluations cover areas from data quality and model-centric metrics to real-world user experience, often employing human feedback alongside automation for comprehensive assessment.

What’s Wrong with the Turing Test Picture?

If we were to roll up all of the problems associated with the Turing Test into one statement; it might read something like this; “mimicry is not the same thing as intelligence.” In other words, how the model’s outputs or responses are produced is just as important, if not more important than the outputs and responses themselves. Let’s take a moment and explore that a bit…

Polly isn’t really intelligent… What if you’ve got the most talented Parrot ever found / trained and it can utter 100 phrases or more in English (or another language, it doesn’t matter which); not only that - this parrot can produce those ‘responses’ based upon contextual cues provided by the owner prompting it (to speak). Here are some relevant questions about Polly…

1. Did Polly learn how to speak? No, not really - it did figure out how to provide the expected responses though.

2. Does Polly understand what it’s saying? No, not at all.

3. Will Polly be able to recall its many conversations and fit them within a Continuity of Interactions with its owner (prompter) that help Polly evolve as a self-aware being? Not really. Polly probably does have a little of this, but nothing approaching a human level - and in this area at least, Polly may actually exceed what current AI models are capable of.

Polly is in fact an excellent metaphor for what’s happening with the current generation of LLMs, (and in fact most if not all AI models). Let’s say we gave Polly a $5 billion Datacenter instead of the pea-sized brain it was born with and who knows, maybe it would be competing successfully with all of those AI models using today’s benchmarks.

This analogy helps us to understand the key problems with both the Turing Test and today’s expectations for what AI is; these problems include:

• The confusion between output and intelligence.

• The inability to properly define our own intelligence; let alone any others.

• The inability to evaluate something that thusfar cannot be properly defined (this is a big one).

It’s worth noting here for a moment that AI isn’t the first time we’ve faced this situation; in a way, nearly every computer-related capability ‘mimics’ us and now it actually typically exceeds human outputs (Chess, Math problems etc., etc.). It’s just that when this shifted over from math-focused computations to ‘creative outputs’ such as writing, images and videos that we’ve suddenly lost all of our previous context for what was actually happening. All automation replaces things that humans do and computers have been reproducing the products of human ‘thought’ for more than 80 years. That didn’t make computers ‘intelligent’ in the same sense we are then and it still isn’t doing that now.

How Do We Fix This Dilemma?

We start as all good problem-solving ought to start by acknowledging our flawed assumptions; in this case, all of those arise from the core assumption that Intelligence should (or can) be conflated with Mimicry and then actually be achieved that way (it can’t). The fault is in ourselves here - it’s our perceptions which are confusing this situation more than any technological constraint or problem; because those perceptions have had us chasing down the wrong solutions. It’s almost as if we’ve been trying to solve a three-dimensional problem with two-dimensional solutions, (the two dimensions being the two sides of the interaction and third missing dimension being those processes involved in self-aware thought).

To really fix this situation, we’ll need to also reimagine the problem (in order to create a new metaphor) while simultaneously solving an age-old riddle; that riddle being “what is intelligence?” This redefinition won’t be easy; if it were, it would have occurred a long time ago and we certainly wouldn’t be in the absurd situation now where people are spending a large % of our GDP to build data centers the size of small cities.

Part 2 of this article series will look at what a new metaphor for (actual) Artificial Intelligence, (and thus the solution / industry foundation), should look like. Part 3 will then address how to evaluate what comes from that new foundation…

Copyright 2026, Digital Perspectives

There is a difference between mechanisms behind Intelligence and the outcomes manifested from it; yet when approaching this from a technological perspective (in terms of trying to recreate either), we tend to confuse the two. There is no such thing as a monolithic intelligence, even within one entity (either organic or artificial), multiple types of intelligence work together synergistically to produce outcomes that we then attribute to a "General" intelligence. However; there is no such thing as a "General Intelligence" and thus trying to build one becomes an impossible task. (thought of the day posted on Linkedin.com, Dec 12, 2024)

Over the past several years, I’ve given a lot of thought to what AGI is (or isn’t) as well as how we might achieve it. It’s one of those situations perhaps where you’re caught in the river of popular consensus (or maybe even a Hype Cycle), and it’s so powerful that you ignore or miss altogether the flaws associated with the trend and / or its related concepts. That’s definitely the case here. It dawned upon me recently that the premise upon which the next generation of AI capability is being developed is fundamentally flawed, which means that it’s likely the entire industry is building upon quicksand. The problem takes us back nearly to square one.

The notion of AGI won’t work, at least not in the near-term, and perhaps never (at least not the way it’s being envisioned now). This means that a whole lot of effort and investment is likely being misdirected. I’m going to tackle this first at the conceptual level with the following articles:

1. Re-Considering AGI (this article)

2. Re-Imagining Intelligence (and thus AI)

3. Re-defining & Replacing AGI

4. Replacing AGI Expectations

First of all; here are some of the previous articles I had written on the topic:

• What are the Special Responsibilities associated with Development of AGI?

• What is the AI Revolution?

• Orders of Artificial General Intelligence (AGI)

Much of what in those articles may still apply, but per the previous comment, the overall premise upon which all of those examinations was occurring is highly problematic. Let’s review just some of the problems associated with the current definition and expectations with what’s been defined to be Artificial General Intelligence:

1. It is sufficiently vague enough (despite attempts by Google, myself and many others to define it), to represent an achievable target.

2. Worse yet, given the myriad of interpretations yet to emerge (and there will be many), it will be even harder to assess and evaluate. This has already been occurring with nearly non-stop claims over the past year that various models / products utilizing GenAI technology had already achieved some form of AGI (reasoning, sentience etc.). Without standard definitions and measurements, anyone can continue to make such claims which in turn makes the entire field more marketing than science-focused.

3. The biggest problem though is that the misdirection in expectations has a direct impact on what development occurs (and how it will occur). In other words, the current confusion and flaws associated with the current definition (or lack thereof) will lead to a lot of wasted effort and less than satisfying results. This is not to say that someone somewhere might stumble onto a solution accidently (we’ve already seen that happen with GenAI itself), but it’s unlikely that this will be the case given the nature of the problem / challenge.

4. If we don’t understand how we’re getting our current results (using GenAI), it’s unlikely that adding more power to those models will magically create something else that we can’t actually define (or agree upon). While many seem to be genuinely surprised that current models are “hitting the wall” in terms of how far they can go - the only real surprise is that they got this far.

5. The current (AGI-based) approach to developing the next level of Artificial Intelligence is completely unsustainable (precisely because of dependence on a range of implausible assumptions, the worst of which is the need for “infinite scalability”).

None of this means that we cannot achieve expectations for higher levels of artificial intelligence (if they are better defined); this is not one of those fear-based discussions about how trying to achieve it will lead to another vaguely defined outcome (The Singularity). Rather, this is an exploration about how we adjust our assumptions and perceptions of what we’re trying to achieve in order to actually get there. It’s also an attempt to gain more control over the process by making it intentional and deliberate in a way that simply isn’t happening right now (but needs to).

more to come…

copyright 2024, Stephen Lahanas