Affecto Oyj : “Artificial Intelligence is not there yet – they say”

Despite the noise artificial intelligence (AI) is creating in media, we have seen only a tip of an iceberg, even though the world is filled with researchers, scientists, companies, universities and other institutions trying to make out what exactly AI is all about.

So, what AI is in the first place?

Artificial intelligence is a computer program trying to imitate human brain. In one way of describing it is to say, it is a computer system able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. Yet - it has just started its' a bit over 60 years- journey, while human brain has been evolving over hundreds of thousands of years.

So why AI is not there?

I attended AI World Summit last week in Amsterdam. One of the AI gurus, computer scientist Stuart Russel from University of California, Berkeley, explained the situation as follows;

First of all, we have gotten the definition of AI wrong: 'intelligent machine that optimize given objective'. Instead, what we want, is 'provable beneficial machines' which connect to underlying preferences we want. He also raised a quote from early days of AI: 'We had better be quite sure that the purpose put into machine is the purpose which we really desire' (Norbert Wiener, 1960).

Second point Mr. Russel made, was about the immaturity of technologies. We are still missing

• deep understanding of languages
• integration of learning with knowledge
• cumulative discovery of concepts, theories, actions.

All these require conceptual breakthroughs, and the date when it will happen is still unpredictable. Should we really make AI and robots work for us for human benefit, their 'objective should be to maximize the realization of human values'. Yet robots are uncertain what those values are, so human behaviour should provide that information. Which leads us to the third concern of Mr. Russel pointed out:

The problem is really us, the people.

The difficulty in benefiting the AI technology underlies in our computational limitations, our inconsistent preferences, internal conflicts and the fact that people are nasty.

Gary Marcus, from New York University (also founder of Geometric Intelligence, acquired by Uber) also claimed that we are not as closely to 'strong artificial intelligence' as many of us believe. Despite the progress made in algorithm development, we still haven't got safe reliable driverless cars, domestic robots, automated medical diagnosis or conversational interfaces. Why is that?

First, he claims machine learning (ML), a subset of AI, uses probabilities to in order to give output for a certain problem. It is hard to debug, difficult to revise incrementally and verify.

Secondly, he claims the statistics does not equal knowledge.

Thirdly, there is a standard bias in the field to assume that virtually everything is learned. I suggest you take a look at this article to clarify Mr. Marcus' thoughts.

Available here

Human brain is too complicated to imitate with current available technologies. One single science cannot understand the function of brain. Mr. Marcus made it very clear that one 'must bring together psychologists, linguists, neuroscientists, and philosophers in order to understand the human mind as a whole'.

Therefore, the development of artificial intelligence requires many other disciplines to work jointly with technology. He further claims that AI should be used helping world as broadly as possible, instead of narrow focus of large corporation's way of making more money via marketing recommendations.

As long as we humans cannot say precisely what we want, machine's reasoning is uncertain. It should be our common goal to share resources among human race and not to generate zero-sum game of AI, where one rule over other. Democratization of AI is on its way via open API's. This should benefit all human kind.

Calling for diversity, Wendy Hall, Professor of Computer Science and Director, Web Science Institute, University of Southampton, raised an important question of girls in computer science. I urge you to read this article to get acquainted with the topic.
Available here

As a summary of academic topics in World AI Summit, I learned the following things;

1. we still have a journey to go in order to call technology truly artificial intelligence
2. it is really multidiscipline science, which calls for team work and diversity
3. the analysis of unstructured data was the most favourable topic in the field
4. large corporation in B2C are currently the drivers of industry
5. the topic of ethics in AI is trending now

This is one of the three blogs in a series of World AI Summit. In my next writing, I will be overview of presentations on how we use AI currently and what kind of use cases are there today we can be proud of. The final third considers the morale and ethics - stay tuned!