The article presented here was the result of a Starttech internal discussion on AI issues. It was translated by the author, Prof. Georgakopoulos, with some minor editing, from the Greek prototype.
- A short prologue
- Logical deduction
- Awareness of actual conditions.
- Problem solving.
- Goal setting and pursuing.
- Assimilation of isolated examples.
- Kinesthetic perception.
- Literature and “figures of speech.”
- Pattern recognition \ Pattern discovery.
- Expansive analogies – similarities and differences.
Speaking about us, humans, we are rather weak as beings of nature. We lack the physical strength or the sensory capabilities of other animal species. Yet we do have a remarkable intelligence, which, as far as I can see, was enough to elevate us to the top of the food-chain on Earth…
In our era, the digital era, the keyholders of this world have classified this advantage to “goods-in-scarcity” and got involved in heavily time-and-money consuming efforts to create an artificial one, to substitute the natural one. (I fear that this may just happen because human intelligence, dignity, and freedom bear strong kin relationships of the first degree.)
During this, possibly short, intermezzo between the tsunami that GPT-3 brought, and the promised forthcoming of GPT-4, we enjoy a window of opportunity to recollect on where our intelligence lies indeed. And from this, to obtain several criteria whether the various faux-bijoux expected to be brought within our sight in the years to come, will be indeed worthy of their name(s).
Let us recall therefore a quite long series of aspects of our (human) natural intelligence. As an author I cannot claim that the list is exhaustive, nonetheless I did make an effort to render it rich enough and not repetitive. The reader has thus a reasonable “intelligence-check-list”: if one faces an AI system one may scan the list and check which facets of intelligence this system does indeed possess or not.
1. Conceptualization: Our thought includes as a central and vital element the “concepts”. [The corresponding Greek word means literally the “in-minds”.] We have not yet formed a sufficiently clear description of what is happening in our minds in order to form “concepts”. Yet we do live “within” them. Imagine that you hear two pieces of news: “The zoo of our town returned its dolphins to the sea,” and “Yesterday I left my parrot to leave its cage”. Don’t you sense at once the presence of the same concept, “freedom,” in both news? What could (or is going to be) the functionally computational equivalent of such an “understanding”?
2. Logical deduction: The capability of discovering logical consequences lies in the core of our mind and intelligence. If John and Maria dislike each other, we can easily deduce that they will take care not to be both at the party of their friend tomorrow. And if we are asked “Who is the mother of the children of Mary”, we can readily answer the question – (in relation also to (1)).
3. Awareness of actual conditions: Side-by-side with the logical machine in our minds, lies our memory – a stunning device, full of the actual experiences we collect from the actual world “out-there.” What is of paramount importance here, is not just this “storage” of facts or knowledge. “Storing” is already happening in any written text. The important fact here is that this memory is continuously co-functioning with the rest of our mental actions – mainly through the mechanism of “connotations” or “(mental) associations”. We are still lacking an effective way of implementing such a mechanism of an “active knowledge base”, equally useful with that we all possess. E.g., if we were to ask a computational system “which reason(s) people may have for preferring olive-oil to sunflower-oil? And which reason(s) for the opposite?” through which sort of “knowledge” would this system be capable of answering?
4. Problem solving: Solving practical (and “theoretical”) problems is of course a central mental capability, one giving to our intelligence a, literally, survival value. We do not refer only to problems like puzzles or logical ones, etc., but to all sorts of them. What is crucial with respect to this issue, as far as intelligence is concerned, is that apart of just solving a given problem; we are also capable of detecting and formulating problems (to be solved, possibly quite later on). E.g., in which way could a digital smart system realize the inconveniences we sometimes meet in putting on and taking off our clothes, and thus set a goal to help on this “problem” by inventing the button, or the zippers, or even the ‘velcro’ tape!
[ Rather recently, (Autumn 2022), several massive demonstrations took place in China, as a protest against unbearably tough quarantine measures. The state, at some point, prohibited the public use of posters or banners on this issue. Well – this is a “problem”! And the people’s response and solution to it was brilliantly ingenious: protesters in the streets were just holding blank A4 sheets of paper! Naturally, everyone was intelligent enough to imagine what was (not) written on them. (Despite several AI systems publicly available at that time, the chance is that Shanghai-people took no such advice from them…)
I am eagerly waiting for the moment at which some dictator (a political clown, that is), will issue an order of the form: “Public waving of blank sheets of paper is strictly prohibited.” – in which case of course the people protesting will just draw a large Χ on each sheet. 🙂 ]
5. Goal setting and pursuing: Action planning towards a specific goal, even a simple one, is another vital aspect of intelligence. Such “goal driven” planning is still considered to be a first class goal for AI. Yet, we should not forget that before any planning, a deeper event has taken place: setting the “goal” itself. In this process actual states-of-affairs are evaluated and values prioritized, by persons or communities, having in mind a scope of days, weeks, years, or even of a time-life-span – until step-by-step whole cultural environments emerge. A robot may set the goal to screw some screws, yet it is not capable of offering you or us a target, especially one of a long time-perspective.
6. Assimilation of isolated examples: Assimilating examples is a basic learning mechanism. “Artificial neural nets” were designed exactly for imitating this mechanism. Yet neural nets, in order to be trained efficiently, need a lot of examples: thousands of them or even millions. What is impressive in the human case is that quite often a single example suffices: we show to some friend how to peel off an apple and subsequently he or she is capable of peeling any apple. In some sense even “zero” examples suffice: our friend will be also capable of peeling off potatoes (not given as an example), or even pears (which do not even have the same spherical shape).
7. Kinesthetic perception: Our intelligence involves not only the mind, but also the body (and the soul), and sometimes coordinates all of them in amazing ways, (like e.g. when one is dancing), sometimes without even using our eyes. This capability allows us to perform a great variety of “tasks”: from tying our shoestrings to playing the guitar or piano, and even balancing a bicycle under quite unfavorable conditions (like, on irregular mountain-ground, or running with 40km/h on a stadium track of 20o slope just-for-sport).
Here: https://www.youtube.com/watch?v=wXxrmussq4E one can see a four-legged robot capable of opening a door. Interesting! Yet, there exists many second thoughts, in order:
1) How did the robot understand that the door was a rotating one (and not a sliding one?)
2) How did it decide the way to rotate the handle of the door? (Since in many doors the handle has a vertical position, and it is rotatable either CW or CCW.)
3) How did it check whether the door was locked or not?
4) How did it judge that the door was opening along the “inside” direction (vs the “outside” direction)?
5) (In which last case, how was it going to take care of any unfortunate guy just standing coincidentally on the other side?)
(Of course, the robot most probably did nothing from the above. As far as I can judge, they were all “hard-coded” from the beginning.)
8. Creativity: Well… this is one of the greatest mysteries with respect to our intelligence: that moments of unexpected leaps forward it is capable of offering to us. Sometimes with great historical impact. (In science these leaps have been called “scientific revolutions” and the landmark analysis of them is the classic, by now, book: “The structure of scientific revolutions”, by Thomas Kuhn.) This phenomenon is of course equally important in art. Suppose that one has trained a “DrawGPT” system with all paintings before 1907. Does anyone hope that such a system would be able to paint Picasso’s “Les Demoiselles d’Avignon”?
Such creative leaps forward are not necessarily always revolutionary. We experience events and actions of micro-creativity almost everyday.
9. Literature and “figures of speech”: NLP, (natural-language-processing), was one of the first targets of AI – one of high priority. Yet, let us not forget that natural languages have many facets: they follow a set of rules, but they also offer a lot of meaningful (!) exceptions to these rules. And we praise not only any “exactness” of speech, but also quite a lot of artistic deviations from it. I do wonder on what principles an NLP system would interpret a proverb? I cite here e.g. the following Greek saying: “A clear sky fears no thunderbolts”. (One can find similar, or even more obscure, sayings in every language easily understood by most people.).
10. Pattern recognition \ Pattern discovery: Pattern recognition was one of the first steps of the discipline of algorithm design. Yet our intelligence is not only capable of recognizing (given) patterns [e.g., “find all square objects in this picture”], but also of discovering new ones: we are able to detect that a new pattern emerges in some state of affairs. E.g., one may draw from the sea a lot of fishes in one’s nets and observe: “ Look! This is the first time we catch this sort of fish! ”
An impressive relative example of pattern discovery, are the puzzles of M.M. Bongard, (originated back to the ’60):
In the figure above there exists a rule that is followed by all icons on the left, and which is not followed by none of the six icons on the right. Do you see it? Sixty years after their introduction we have not even started, in some sense, to design systems capable of managing such issues. (The author is aware of some relevant research along directions as the above one, yet he also knows a case where the researcher abandoned the study of AI, declaring his inability to exercise any control on its misuse – namely for military applications.)
Our wider intelligent capability of a “state-of-affairs-awareness” is not confined to logical state-of-affairs. It is expanded along wider horizons of emotional or ethical “intelligence” – issues having for us equally grave importance and impact.
11. Expansive analogies – similarities and differences: It is not a secret that the use of analogies and similarities play a crucial role in our intelligence. What is interesting here is that we possess the extra capability to “extend” them to cases where the analogies break. Sometimes we may do so by paying attention also to the differences and not only the similarities. E.g., the system AlphaZero plays chess impressively well. Yet it becomes instantly useless if we agree to change or bend a rule, even slightly, for example if we agree to restrict the movements of a bishop to, at most, 3 squares at a time. On the other hand, two good chess players are able to play such games- just for fun. (On this issue one may want to check the suis-generis culture of “fairy chess”: https://en.wikipedia.org/wiki/Fairy_chess ).
[ A friend of mine, an excellent player of preferans, became a strong and winning player of bridge during the first game of bridge he played, having just been informed about its rules – (against amateur, yet quite experienced opponents). ]
12. “Proofs”: “Proofs” form a core element of mathematics, and belong to the category of “logical deduction(s)”; yet they deserve a special mention for three reasons: (1) in many cases they involve a logical analysis towards proving or disproving of specific statement (the formulation of which challenges frequently all of our aforementioned capabilities); (2) while their structure has purely syntactic characteristics, they are achieved in a way that is tightly interwoven with their semantic content; (3) they refer to “mental” objects lying in most cases quite beyond our immediate experience, (hence: where these semantics may come from !? ).
Notice that we are all more-or-less “mathematicians”, as we are all more-or-less painters, (although not necessarily at the level of the most brilliant of them).
Jokes and “games-with-words”: Last-but-not-least we should mention that our intelligence produces and (greatly) enjoys humor. It may laugh. (From the rest of the animal kingdom some sounds of the dolphins may be laughter. Monkeys also seem to laugh and sometimes they even wave their hands in a way reminding of similar human gestures during a strong a loud laughter. )
Many “jokes” have a practical form, but most of the time we “joke” in oral and written speech. Check the following two phrases, the 1st from a well-known book of J. Hašek, and the 2nd from a cartoon of Greek cartoonist Y. Ioannou, (about cities’ smog):
– “The court decided that the complete lack of any accusing evidence against the accused is against him, since it reveals that he took conscious and careful care for eliminating all of it.”
– “Occurrences of this phenomenon are exceedingly rare. Frequent recurrences, even more.”
What can we say to, or be said to us by, a computer system in such a “language”?
S. Freud had devoted a monograph on humoristic expressions supporting the view that the main reason for their pleasing effect on us, is that they achieve a very high density of meaning. (This reminds us of “data compression” yet Freud meant more than that.)
It is exactly this multi-facet natural intelligence of ours that we should preserve, cultivate and take full advantage of, on any road leading possibly towards prosperity and civilization worthy of their names.
Unfortunately, in our era, one can find many people, (technically talented, but perhaps naive or even greedy), believing that a sufficient condition for progress is three-fold: themselves, a budget from info-tycoons, and the beans of Java or Python.
Fortunately, or unfortunately, this is not the case. Our intelligence is quite powerful to provide us with sufficient advice on this issue…
PhD on algorithms and complexity,
Faculty member, University of Crete, Greece.
Prof. Georgakopoulos began its academic career as a PhD student of Christos Papadimitriou at NTUA, and since 1993 is a faculty member of CSD at the University of Crete, Greece.
Currently (Spring 2023) he is on a sabbatical leave, working with Starttech Ventures on Algorithms, Complexity and Neural Networks.