Artificial Intelligence – a “copilot” for the 21st century

Artificial Intelligence (AI) is the hype of the moment, largely due to the significant advances of some generative applications, notably ChatGPT, whose language model is truly impressive.

Alluding a little to fiction, computers are finally beginning to resemble HAL9000, the protagonist of Stanley Kubrick’s “2001 – Space Odyssey”, based on a short story by Arthur C. Clark, who co-wrote the screenplay. However, AI and its foundations go back a long way.

In 1950, Alan Turing, the English mathematician, philosophized about the thinking of machines and their intelligence. Then, based on the dynamics of the “imitation game”, he devised a test in which an evaluator talks to a computer and a human (without seeing them directly). If the evaluator was unable to distinguish between the human and the machine, this would mean that the machine was “intelligent”. This test became known as the Turing Test and, conceptually, is still valid today.

However, even with all the advances made in the meantime, AI is still a statistical engine that is only as good (or as biased) as the training it has been given. In other words, the machine reproduces what it has been taught. If, absurdly, it had never seen a picture of a cat, the machine would never be able to discover a cat[1]. In fact, much of the progress made in AI today is due to the huge library of knowledge and examples called Internet. It is this huge database, the largest in the history of mankind, that allows algorithms to be trained so well. And that’s exactly why the machine is sometimes biased – because so is the sample of experimentation and training. If we train the machine with our prejudices (albeit unintentionally), in other words, with concepts and formatting, often with a lack of bias, the result is a system formatted with biased concepts. But let’s be optimistic, because the truth is that “the machine”, when well trained, produces a very efficient decision-making system and makes AI a very powerful tool, capable of drastically increasing our productivity.

 

“Listening” water leaks – a good example of applied AI

One of the most interesting practical examples I know of AI’s ability to play a major role was created by an ARQUILED partner, the Brazilian start-up Stattus4, to detect water leaks in the public water supply network, with the goal of reducing the enormous waste of water that continues to occur.

Typically, water pipes or plumbing are buried underground, so it is very difficult and costly to detect leaks early on, because periodically digging up the pipe for inspection is not feasible. To overcome this difficulty, Stattus4 has developed an AI system that, based on 10 seconds of sound captured in the pipe, can detect whether it is leaking or not.

And the way it does this is extremely simple. Experience tells us that a pipe with a hole through which water pours has a certain vibration, which is matched by a sound “signature”. Competent professionals with keen ears can detect this distinctive mark, but a person without this preparation and experience cannot.

What Stattus4 has done is to use AI to “train” and multiply “keen ears” without needing them. In just a few minutes, the system can classify thousands of sound samples from potentially leaking pipes, which would be humanly impossible without a battalion of specialists.

With this tool, an experienced professional only must validate the suspected points of leakage that are identified. In a scenario where the AI system has classified 1% of the samples as suspicious points, the (human) professional only needs to take action to validate the machine’s classification. But he only must analyze 1% of the samples! In other words, a precious and highly qualified resource can increase the productivity 100-fold.

I think this example demonstrates just how powerful AI can be when it is widely applied. By filtering out large amounts of irrelevant information, it lightens the human’s load, increasing productivity and allowing humans to focus on where their knowledge is really needed. It’s no coincidence that Microsoft calls its AI system “Copilot”. That’s the idea: to have copilots who help us to be more productive.

 

Miguel Allen Lima
ARQUILED CEO

 

[1] We could debate whether the same applies to us humans, but that would be the subject of another, much longer article.