AI: reasoning using data?

This is a text automatically translated from Italian. If you appreciate our work and if you like reading it in your language, consider a donation to allow us to continue doing it and improving it.

The articles of Cassandra Crossing I'm under license CC BY-SA 4.0 | Cassandra Crossing is a column created by Marco Calamari with the “nom de plume” of Cassandra, born in 2005.

Cassandra explains a little about how AI works: can they really think for humans?

This article was written on 06 January 2022 from Cassandra

Cassandra Crossing/ AI: reasoning using data?

Artificial intelligence techniques are more often abused than used

As Cassandra's 24 graying readers know well, Artificial Intelligence and its applications have been the subject of Cassandra's thoughts for a long time.

Now that they are among us, the mood of your favorite prophetess alternates catastrophic perceptions a la Skynet with reminiscences of K.'s misadventures between Trial and Castle.

But the real responsibility for the abuse of the term, and especially for the improper use of Artificial Intelligence must be divided equally into three parts.

First of all, research has been studying the question for almost a century (since 1956 to be precise), obtaining preliminary results and promising wonderful things in the near future. The public becomes familiar with such a catchy term, and as we know, it does not always distinguish imagination from reality.

Secondly (do you like macaronic Latin?) because the narrative, starting from the myth and the novel to reach the big screen, the small screen and the Internet, has made it a subject as beautiful as it is interesting, and we know, in the minds of People imagination and reality coexist at all times.
Thirdly, because talking in buzzwords and selling rubbish has always been a source of money, power and reputation.

And so when, after Inference Engines and Neural Networks, Deep Learning, a new application of artificial intelligence, started to produce interesting results in a simple way, everyone jumped in.

And if the narrative has benefited further, the practical application has begun to be stretched and extended into all possible fields, with the hope of generating new business.

And this is a very dangerous blunder.

Deep Learning, and for that matter Neural Networks, have nothing to do with “Intelligence”.

“True” intelligence, which artificial intelligence researchers call “Generic Artificial Intelligence” may not require self-awareness (we leave the question to philosophers) but it certainly requires, in addition to knowledge, both understanding and logic.

And Neural Networks, like Deep Learning algorithms, certainly lack both.

Now, if feeding (with a lot, a lot, a lot of effort) the Inference Engines with "rules" certainly makes them "logical" problem solvers, because they can do "backtracking", that is, explain how they arrived at a certain conclusion, the algorithms of Deep Learning can only tell you if a photo represents a kitten or not, and this only after being fed many photos of kittens and non-kittens, with the solution written behind them.

But they can't tell you the "why"; they do not apply intelligence, they do not know "Kitness", nor other logical categories.

Therefore, if it is useful and reasonable to apply Deep Learning systems in suitable issues, even very important ones, such as support for tumor reporting, it is absurd and dangerous to apply them in fields where "true intelligence" is necessary, deluding oneself into thinking that it can replace it and also obtain more accurate and more economical results.

Court sentences, personnel selection interviews, the possibility of recurrence of a crime are issues "evidently" outside the capabilities of a Deep Learning system, even if unfortunately it is possible, indeed easy, to build a "fake" one that can be used in these sectors.

And the world is unfortunately full of sellers, perhaps even partially or totally in good faith, of snake oil.
Who controls the results anyway? Who can say if the “criminal” really repented?
From the data, and among other things only from the "good" ones, we can and must learn, but as we have always done.

Data, and only good data, can be categorized logically and offer ideas for better categorizations, allowing us to discover "data within other data".

But data cannot replace intelligence, data cannot "reason" in place of humans, or at least those humans who bother to do so.

Marco Calamari

Join communities