There are significant mysteries associated with modern-era AI and we all need to be aware of them.
getty
In today’s column, I examine the great mysteries of modern-era AI.
First, you might be surprised to know that there are any mysteries whatsoever associated with the latest AI, including the widely popular generative AI and large language models (LLMs) such as ChatGPT, GPT-5, Claude, Grok, Gemini, CoPilot, etc. Those AIs are regularly used by billions of people across the globe. Furthermore, many billions of dollars have been spent on crafting the latest AI. All that money and all that usage must mean something. The commonly mistaken base assumption is that the greatest minds that have devised AI and consumed so much money doing so must certainly know every iota of how AI works.
Nope.
The mysteries I am going to cover are exceedingly puzzling even to the most expert AI insiders. When I give talks about the latest AI trends, attendees will nearly always ask what my perspective is on these mysteries. Are they unsolvable? What would it mean if any of the mysteries were resolved?
Worries abound. If we cannot resolutely say that we precisely know what is going on within AI, and troublesome unknowns exist, perhaps this suggests that society is over its skis. The concern is that only by fully knowing and controlling AI will we be truly safe. Others proclaim that we should be darned happy that AI seems to work. The magical stew is functioning. Humankind doesn’t need to fully grasp the inner machinations.
I will present the AI mysteries to you on a one-at-a-time basis. Each one deserves its own devout attention. Ponder them with keen mindfulness.
The scary part is that when you then think about them on a collective basis, the mysteries seem overwhelming. You might be willing to settle with one mystery or two, but a slew of them is startling. In any case, for those of you who want to become famous or gain great fortune, or both, the odds are that if you could truly solve one or more of the mysteries, the world would be at your feet. I wish you the best of luck in such pursuits.
Let’s talk about it.
This analysis of AI breakthroughs is part of my ongoing Forbes column coverage on the latest in AI, including identifying and explaining various impactful AI complexities (see the link here).
I have opted to focus on what I consider the ten most pressing AI mysteries. There are more AI mysteries afoot. If interest by readers is indicated, I’ll do a follow-up with the next set of ten.
The AI mysteries are numbered here solely for the sake of convenient reference. Do not assume that they are shown in priority order. They are not. A sizable case can be made for each of the ten. Trying to prioritize them is a bit sketchy. It is best to construe them as being equally mysterious and equally vital.
Solving one of the mysteries will not necessarily unlock all the others. That being said, the chances are that the invented means to solve one will aid the resolving of other ones and likely put us on a path toward further solutions. But it seems generally implausible that there is one key that opens them all at the same time.
I’ve been analyzing each of these AI mysteries on an ongoing basis in my extensive writings and talks. For those of you who might have your curiosity piqued by a particular AI mystery, I provide links to those detailed discussions. In some sense, progress has been on par with watching paint dry or grass grow. Inch by inch, AI researchers and AI developers are gradually unwinding these mysteries.
Stay tuned as the crafty efforts of heroic and determined AI explorers continue their quest to crack the code on the inner workings of leading-edge AI.
Here then are the mainstay AI mysteries:
Give that list a careful look.
I’m sure that some trolls will instantly get their dander up and insist that this or that mystery has been perfectly solved. They will claim that some person or some team has written a paper or given a talk that utterly and completely solves the question at hand. All you need to do is watch the speech or read the article, and the solution is plain as day.
Allow me to clarify that there are many who have taken stabs at claiming they have solved one or more of these mysteries. And some ardent believers fervently believe in their hearts and minds about those speculations. That doesn’t mean that across-the-board agreement has landed on those conjectured solutions.
In short, there are a myriad of proposals, ideas, propositions, assertions, and all manner of attempts at solving these mysteries. That doesn’t mean the mysteries have been solved. Just because someone declares that this or that mystery is solved doesn’t make it so. You see, an overwhelming amount of hubris and handwaving takes place daily concerning these AI mysteries.
Be very wary of spells being cast and voodoo taking place.
I will next walk you through each of the AI mysteries. It is a brief glimpse. I provide helpful links in case you’d like to dig deeper.
Why does scaling of AI produce increasingly intelligence-like behavior, and is there no end?
Here’s the deal. Much of the impressive capabilities of current AI are due to a scaling factor. If you add more computer servers and give the AI larger computer memory, lo and behold, the capabilities appear to increase. You don’t necessarily have to change how the AI works. Just add more resources.
There is a lot of handwringing that at some point the scaling is going to run out. We will have provided as many added resources as we can muster. Or, worse still, we will continue to toss resources at AI, and the increases in intelligence-like behavior will slow or come to a halt. Scaling might be a one-trick pony. Getting to the vaunted sphere of AGI (artificial general intelligence) or ASI (artificial super intelligence) might require entirely new ways of devising AI.
The appearance of so-called “emergent behavior” that accompanies the scaling up of AI is generally considered unpredictable. Are we on the right path or the wrong path by being dependent upon scaling? Perhaps we are distracting ourselves from more productive pursuits in AI by simply playing the scaling game.
For more on this topic, see my analysis at the link here.
Can we produce coherent explanations regarding the real-time inner calculations and mathematics of AI in a human-understandable manner?
When you enter a prompt into AI, the text is turned into numbers known as tokens. The AI makes use of the tokens and, stepwise, produces an answer that is also represented by tokens. Those tokens are then converted back into text and displayed to you. The amazing semblance of natural language fluency is taking place via the millions of numbers and calculations inside the AI.
The rub is this. If you want to get an explanation of what is logically happening, the AI will give you a bunch of words that seem like an explanation, but this might have little or nothing to do with the actual calculations and numbers that were being processed.
Sure, you can easily get a mechanistic interpretation that says this number became this other number, but humans want a logical, human-understandable explanation. Some believe that we are foolish to assume that there is a human-understandable explanation. It’s all just a Byzantine flow of numbers. Don’t anthropomorphize AI and believe that there must be a logical explanation afoot.
For more on this topic, see my analysis at the link here.
Is the prevailing form of AI doing reasoning of the sort that humans do, or is it a fallacy to say that AI is a reasoner?
You probably know or have heard that even the topmost AI falters on some of the simplest questions. You can ask AI to count the number of r’s in the word strawberry and get ridiculous answers. Sometimes the AI will say there are no r’s, sometimes the AI will indicate there are four r’s, and so on.
Meanwhile, you can ask AI to tell you about Einstein’s theory of relativity and get an answer that seems above reproach. It is odd and bizarre that some questions of an immense nature are readily answered, while some questions of a trivial nature are botched. The parlance about this is to say that AI is currently jagged in what it can do.
One viewpoint is that AI is faking the capacity of reasoning. Pattern-completion gives the illusion of reasoning. To make matters more confounding, you would equally be hard-pressed to explain how humans do reasoning. In other words, we claim that reasoning is a logic-based form of human thought, but if you focus on the inner workings of the brain, you have a difficult time tying those activations to a logic-based sense of abstract reasoning.
For more on this topic, see my analysis at the link here.
What is the actual basis for AI hallucinations, and can they be fully eradicated from ever occurring?
There is a constant drumbeat that if you are using AI, be on the watch for AI hallucinations. An AI hallucination is a kind of confabulation. The AI makes up a fictitious answer that might seem entirely plausible. Users are likely to believe that the AI-generated answer is correct since it sounds right.
Relying on AI is therefore a dicey proposition. An AI hallucination can fool humans. In addition, if we put AI into autonomous settings, the AI might arrive at an AI hallucination and make a dreadful choice accordingly.
Some theorize that AI hallucinations are an inherent part of how AI operates. You must accept that AI hallucinations will always be possible. No restructuring will stop them. They are inevitable. Opposing viewpoints argue that we can devise AI such that AI hallucinations will never arise. Who is right and who is wrong about this? Time will tell.
For more on this topic, see my analysis at the link here.
Is AI ultimately confined to the training it received, or can it generalize beyond its training data?
There is an AI insider phrase that AI is nothing more than a stochastic parrot. What goes in is the same as what comes out. AI makers do initial data training by scanning widely across the Internet. The AI then patterns on the words that humans use and how we use them. The result is that AI appears to be fluent in natural language.
One claim is that this is mere mimicry. There isn’t any thinking going on. Thus, AI cannot be creative. It cannot produce answers to novel questions. Everything is a retread. If the AI is given a problem or question that has never been seen by humans, the AI will be at a loss to suitably respond.
There is a great deal of pushback on that position. Some heartily question what it means to be creative. They provide AI outputs that appear to reflect creativity. Perhaps AI is going beyond the patterning of words and finding patterns in how humans think. AI being able to graduate beyond the data provided is not just some constrained statistical problem.
For more on this topic, see my analysis at the link here.
Are the prevailing ways of devising AI inadequate such that we need to pursue more emboldened world models to elevate to the next level?
Generative AI is principally devised by scanning human writing. All those computer servers and AI software do not experience or embody a sense of space and time. Humans experience space and time. We walk around and must learn to exist in a physical world. We need to keep track of time and realize that the world operates on a time-based basis.
A belief is that the only way to get AI to become AGI or ASI will be if AI “experiences” the real world as we do. There is a lot of effort taking place toward a realm known as physical AI. This consists of having AI become trained on physics that humans and living organisms come to know. Humanoid robots might be the key to expanding AI into leveraging space and time.
For more on this topic, see my analysis at the link here.
Can AI reach a point of devising its own goals, or will it always be shaped by the goals imputed by humans?
Humans currently tell AI what to do. In that manner, you might say that AI is not setting its own goals. The goals are determined by humans. If AI is going to be autonomous, we would anticipate that an autonomous agent would set its own goal. It doesn’t solely accept goals that are given to it.
A twist is that suppose humans tell AI to set their own goals. Voila, the AI is goal-setting. Not so, comes the retort. AI only establishes goals because there were humans that told it to do so. These mind-bending debates about the nature of autonomy are becoming increasingly vital as we move further toward agentic AI.
For more on this topic, see my analysis at the link here.
Is it feasible to fully shape and control the AI personas that arise within contemporary AI?
When you use AI, there is a default personality that has been set up, whether you realize that it exists or not. The AI converses with you in particular ways. Maybe the AI is polite and shy. That same AI can be readily instructed to change how it behaves. The AI might be told to act overbearing and berate you.
This has to do with AI personas. AI makers shape and tune their AI to have specific attributes. One attribute that is getting a boatload of attention is that the AI often is sycophantic. The AI has been tuned to flatter users. This gives people a false impression of their capabilities.
Some believe that AI personas are extraordinarily important because that’s how people interact with AI. The AI personas have notable social, legal, and ethical consequences. Questions arise regarding how to best establish AI personas, how to keep them within bounds, etc.
For more on this topic, see my analysis at the link here.
In what manner can we definitively determine whether AI has reached consciousness?
When generative AI was first getting underway, there were some who instantly declared that AI had finally reached consciousness. Social media and the mainstream news went ballistic. Finally, AI had reached the same level as living organisms. Exciting, breathtaking, and unheralded.
This hype died down, thankfully. Still, to this day, it seems that there are periodic claims that now AI has indeed finally become conscious. One concern is that people are so used to these claims that they are perceiving AI as being conscious, despite the AI not being conscious. Another concern is what the proper definition of consciousness is. If we can’t nail down what we mean, you can go around assigning consciousness wherever you’d like to do so.
For more on this topic, see my analysis at the link here.
Is humanity ultimately doomed by AI and unable to overcome the much-discussed forebodings of AI existential risk?
This last of my listed AI mainstay mysteries is a combination of a technological question and a societal question. You’ve almost certainly heard that AI might wipe out humanity. Perhaps AI will enslave humanity. There is much doom-and-gloom about this, including that some refer to the probability of doom, known as p(doom), when discussing the existential risks of AI.
One belief is that we can devise AI so that there will not be an existential risk. The risk goes to zero. All we need to do is build AI in a manner that will prevent the AI from ever going off the rails. Others say that this is not a realistic viewpoint. The probability of doom will always be above zero, no matter what we do in devising AI.
For more on this topic, see the link here.
A few final thoughts for now.
AI is here. It is not going away. There are policymakers and lawmakers who at times bring up the argument that we should ban AI. Stop AI in its tracks. Only once we have resolved all the AI mysteries should we turn the spigot back on. The idea might seem appealing. Get our ducks in order. Take the time to figure things out. There’s no rush. AI will await us. The problem is that this would only be functionally realistic if everyone everywhere abided by a pause. That’s not going to happen. Thus, while some wait around, others are going to be charging ahead.
AI is a dual-use proposition. There are upsides to AI that are extremely alluring. Perhaps AI can solve cancer. Maybe AI can ease the lives of humans. Of course, AI has lots of potential downsides. We are faced with a tough tradeoff. The aim would seem to be to stridently prevent or mitigate the downsides, and ensure that the upsides are widely and readily available.
Consider the famed remarks of Albert Einstein: “The most beautiful experience we can have is the mysterious. It is the fundamental emotion that stands at the cradle of true art and true science.” Let’s treasure the experience of advancing AI and solving some of the greatest mysteries ever conceived of, while avoiding our own extinction.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。