3rd wave AI tools evolve for solving real world problems

Statistical models are driving today’s wave of artificial intelligence. But this second AI wave creates its own decision models which are pretty much black boxes. So, what’s up for the third wave? Transparent tools for solving real world problems.

Machine learning is evolving and adapting to customer needs.

It all started back in the 80’s and 90’s with rule-based algorithms. That is, you program computers with certain rules and feed them with data. Out come decisions or predictions based on these rules – like from a very sophisticated calculator.

Now, we are somewhere in the middle of the second AI wave, often named the statistical learning wave.

Black box AI

Here we have very sophisticated models that even skilled technicians and mathematicians can not describe exactly how they work. The general principle is that big data trains statistical models which produces classifiers in the format of algorithms. Next step, you feed these algorithms with data and they produce stuff you can apply for decision making.

“The nasty part is that these models are really complicated and you cannot look into how the output has been produced. In many cases, that is simply not something you can live with,” says says Ulf Schönenberg, head of data science in the Unbelievable Machine Company (*UM), a part of the Basefarm Group.

“Certainly surgeons would not put you under the knife without knowing why a computer has suggested them to do so. Also, top financial management would be very reluctant to base decisions without knowing on what grounds computer analysis is made,” Schöneberg says.

So, does this imply that we should take our hands off AI tools and await the next wave? Not really.  Being head of a department with highly skilled professionals, every day Schönenberg faces very real tasks to solve. And solves them.

Machine learning, not AI

“But, really, we are reluctant to call what we work on as artificial intelligence, but prefer machine learning. AI as of today is very much a marketing term. When machines can transfer learning from one area to another, then we are talking AI,” he says.

How to get there? According to Schönenberg, the path has so far not revealed itself. Much work remains to be done.

Which is also the case with the next and third wave of AI.

“We cannot say how it will be, but we can describe what it should be and that we will eventually get there. The third wave is in market demand and therefore lots of brilliant people will eventually take us there.”

Whitebox AI

One aspect of the third wave has even got a name: Whitebox AI, another expression probably formed by marketers (they seem to be everywhere!).

“Transparent AI is a better description,” the data scientist states. “When we can look into the models and understand how they work, we can use the tool for real world problems. We simply want to know why the algorithms suggest a solution.”

Meanwhile, as humans so far have not discovered the secret of time travel, we are pretty much stuck in the midst of the second AI wave (that is, the marketing expression). So how can Schönenberg and his team produce anything at all?

“Well, if I suggest my people to do AI magic, they will laugh at me. They have these heavy technical degrees and a matching level of professional integrity.

“Still, there are lots of ways. As a start, you need strict control and feed your big data lakes with data that is well-documented and high-quality. Read more about data lakes. Then, you know what the statistical models are chewing and can better trust the algorithm classifiers they produce.”

LIME

Also, tools for understanding machine learning of our time are brought to market. Local Interpretable Model-Agnostic Explanations (LIME) is one technique used to explain predictions of machine learning classifiers and evaluate its usefulness.

“You can merely use LIME to verify to what extent and within which margins of failure the classifiers work.”

So, when can you truly trust machine learning?

From data to rules

– and the other way around

“An ideal algorithm will be where you can feed in rules and data which can create new rule sets which fits in with the data – or the other way around. It will be explainable and work in both directions – from data to rules and vice versa. But, that’s for the third wave AI, in marketing terms,” says Schönenberg.