This is the next step to which artificial intelligence wants to take us. And although we can not yet say that the rhetoric of a machine is superior to that of a human, the path is getting shorter and shorter. I know this sounds quite strange, but this is true.
An IBM computer played a human debate Monday in San Francisco, in which he listened to the arguments of other speakers, refuted them and surprised the audience in what is considered an achievement in artificial intelligence.
With ” hundreds of millions” of documents at their disposal, most of the newspaper articles and academic journals, IBM’s ” Project Debater” (as the device is called) could respond without knowing the subject in advance.
It was not free of errors, true, but the public recognized the merit.
While the human presenters did better on a general level, the machine managed to elaborate more weighty arguments. The computer took praise from its adversaries. Noa Ovadia, national champion of Israel’s debate in 2016, was one of the speakers who faced ” Project Debater “.
“I think eventually, when it can improve what we do (humans) will be something great to make informed decisions, to vote in an informed way, for everything!” He told the BBC.
And that is, says IBM, its goal: to get humans to make quick decisions with as much information as possible.
The machine participated in two debates. The first was about whether there should be more space research financed with public funds. The second dealt with whether more should be invested in technologies with medical applications.
Each participant had four minutes to present the first argument and then a replication shift of another four minutes.
To conclude, they were granted two more minutes.
To give their answers, the computer did not resort to the internet but to a database with sources carefully selected by IBM researchers.
Some answers were really ingenious and could be said to be “human”.
For example, when Ovadia argued that money should be spent on things more necessary than space travel, the computer said the following:
“It is very easy to say that there are more important things to spend money on, and I do not dispute this. No one is claiming that this is the only item on our expense list. But that’s beside the point”.
None of the participants knew in advance what the discussion was going to be about, but it had limited a hundred varied topics.
In an interview with the BBC, Arvind Krishna, director of research at IBM, said that this experiment pushes the limits of artificial intelligence that Google has not yet reached.
” Project Debater tries to master language and language is a matter of nuances,” he said.
Some faults of the machine were, for example, that he repeated the same arguments on different occasions but with different words or that he offered answers that did not have much to do with the subject in question, perhaps the result of a mess in the storage system.
Professor Chris Reed, from the University of Dundee, United Kingdom, valued the demonstration.
“It’s a pretty significant breakthrough,” said the academic, who is not affiliated with IBM.
“Addressing something like debating is not easy. One must be able to resolve many conflicts and then reach a resolution. ”
The research director of IBM also considered that what ” Project Debater ” had demonstrated has a clear commercial value.
“When you carry out your work, you often have to make decisions in an improvised manner and sometimes those who advise you have prejudices or interests.”
“If you can draw on ‘ Project Debater ‘ so that you explain their pros and cons on an issue, you can see both equally and that can lead to a much more useful decision.”
Professor Reed is quite in agreement with this argument.
“I think it would be useful to do mixed teams where both humans and computers participate in the debate and in decision-making.”
“Each one has different strengths and weaknesses, and the idea is to design these teams in such a way that they can work better than if they were only composed by humans.”
But something that can start headaches and cause conflict, said the academic, is to decide what are their sources and what their selection criteria.