June 16, 2023

 

There are predictions and predictions, by Howard Rankin

Deliberate action requires continual prediction of consequences, by Michael Hentschel

 

Ai4 INDUSTRY’S LEADING ARTIFICIAL INTELLIGENCE CONFERENCE

AUGUST 7-9, 2023, MGM GRAND, LAS VEGAS

The Track Record Account has reached new highs!!

TRADING PERFORMANCE RESULTS OF THE INTUALITYAI SYSTEM HAVE MANY INHERENT LIMITATIONS.  NO REPRESENTATION IS BEING MADE THAT ANY ACCOUNT WILL OR IS LIKELY TO ACHIEVE PROFITS OR LOSSES SIMILAR TO THOSE SHOWN. IN FACT, THERE ARE FREQUENTLY SHARP DIFFERENCES BETWEEN REPORTED PERFORMANCE RESULTS AND RESULTS SUBSEQUENTLY ACHIEVED BY THE SYSTEM OR PORTFOLIO. THERE ARE NUMEROUS OTHER FACTORS RELATED TO THE MARKETS IN GENERAL OR TO THE IMPLEMENTATION OF THE SYSTEM WHICH CANNOT BE FULLY ACCOUNTED FOR IN THE PREPARATION OF REPORTED PERFORMANCE RESULTS AND ALL OF WHICH CAN ADVERSELY AFFECT ACTUAL TRADING RESULTS.

There are predictions and predictions

Did you observe the great example: of the “hot hand fallacy” during Monday Night Football when the Dallas Cowboys played the Tampa Bay Buccaneers? The announcer Joe Buck mentioned that Tom Brady had never thrown a red-zone interception in his extensive playoff career. A few minutes later, Brady did just that. It was also an endorsement for good old “memory decay” and the fact that a history of previous events probably is not going to predict what might happen on this particular occasion, right now.

Of course, there are predictions and predictions: Is the quarterback going to throw an interception in this game? There’s a lot of information available about this situation. Who are they playing, where and when. What are the weather conditions like? Which team has important players missing? This prediction is “bounded” by the mass of information available.

Now, suppose the question is: “will this quarterback throw an interception on the 10th game of next season?” The information about this distant probability is seriously reduced, making it open to an almost infinite range of possibilities.

Now AI does well in a very “bounded” system: where the possibilities are limited. For example, completing a sentence is relatively easy because both the first few words of the sentence and the specific rules of language reduce the options significantly.

What are the next few words of this sentence opening?

“Hello, Joe…..????”

Something to the effect of “how are you?” is a pretty likely and thus not difficult to predict.

What about the next few words in this opening?

“I want to talk about…..”

Without any other information about the conversation, the people involved, their relation to each other, their conversational history, the answer could be almost anything.

Thus, the word “prediction” can convey different meanings: because the circumstances and information about the situation can vary enormously.

In a relatively bounded system, there is much known: about context and “rules” and a mass of data is going to give you a pretty good idea of what is about to happen.

But where the data is seriously incomplete, unknown and unreliable: – a serious problem for conventional AI – prediction success is not very likely. And mastering this gap will increase the power of AI, taking it to a new level, just as Intuality AI does.

by Howard Rankin PhD, Science Director, psychology and cognitive neuroscience

Deliberate action requires continual prediction of consequences

Adaptive prediction in small probability increments: not oracular determinism, yields more accurate outcomes. The typical computer or human prediction spots a past trend and then states a deterministic assumption that a future event or condition will happen, hoping that the prediction will be true, or at least useful. Typical predictions are not held accountable, nor is there any continuous adaptation to actual future events as they happen.

Real time data and AI are beginning to make progress: Weather forecasting has become more adaptive as real-time sensors give greater clues of what is actually changing. Stock trading with real-time information also tries to keep up with changes to spot counter-trends that might change a prior prediction. But most computer and human predictions are still like the Greek Oracles: predictive pronouncements are made that will either happen or not, without a precise future, and involving no further clarification as time goes on.

Humanized AI innovations can be taught to make more reliable predictions: The key is “adaptive prediction.” Adaptiveness, via real-time reaction to changes in the world, is how humans make value decisions in real life. In a sense, continual evolution is a fact of life, though this is not the permanent kind of evolution that theorists have used to explain changes in the past. Small incremental micro-evolutions are the most accepted reality, and any computerization of probabilistic future reality should take micro and nano-changes into account wherever possible. That requires quite a bit of technology and training.

Humanized AI continually predicts and continually adapts: The current AI excitement is about GPT’s (language processing AI models), which are essentially rules-based, well-informed text/conversation generators. These can however cloud human logic and misinform and disinform in convincing deep fake contexts. GPT’s essentially predict the next best words and factoids in a sentence to create a smooth sequence of predictions that are for most observers equal to or better than normal conversational content, and even sound like coherent logical or scientific debates. Combined with deep-fake avatars, complete videos can simulate/humanize almost anything. Text and data-based simulations are still a relatively simplistic form of rules-based prediction, however much AI is used.

In contrast, Shifting Rules make up the real world in real time: and need to be factored in when predicting real value-based monetizable events where every decision matters in milliseconds. Changing rules open up a system to becoming “unbounded” which is to say even more unpredictable. Tracking both shifting events and shifting rules adds dimensions of interpreted and potentially new patterns that can affect the usefulness of predictions. But growing awareness and tools like IntualityAI for changing patterns and rules can trigger valuable adaptations.

Adding value to predictions requires Action: In IntualityAI’s stock trading applications in the past, one of the uniquely successful features has been its decisions to do nothing (action is doing nothing). The system's 'intuition' is tracked to prevent action merely for the sake of action. It’s important to distinguish actionable events. And then even more, to predict and assess consequences.

Deliberate action requires continual prediction of consequences:  Only in this humanized AI way can human behavior be enhanced without having to make mistakes just to learn failures over and over again. 

by Michael Hentschel, anthropologist, economist, venture capitalist

We'd love to have your comments sent to Reply@IntualityAI.com. All comments are confidential.

This content is not for publication

©Intuality Inc 2022-2024 ALL RIGHTS RESERVED