When I discuss with people about how they perceive Artificial Intelligence and the impacts they envisage it having on their daily lives, I am confronted with the obvious fact that there is a huge pitfall for technology adoption, especially from young adults: the fear of losing their ability to make decisions by themselves.

This article will review a couple of concrete applications to decide whether AI will take away our freedom of thought.

Cancer study at Strasbourg’s CHU

In 2015, the Strasbourg research hospital studied 17 000 anonymized breast cancer cases, in collaboration with a Big Data consulting company named Quantmetry[1], digitalizing millions of handwritten documents and prescriptions, applying Natural Language Processing on them, and then performing data analysis. The developed AI established a clear correlation between hyperthyroidism or type II diabetes and breast cancer. This correlation led to a number of medical publications, and even recommended adaptations in the treatment protocols for breast cancer that had been in use for more than 10 years (in specific rare cases, it made the choice between two possible treatments. Former protocol recommended an alternative treatment that the AI demonstrated to not be the most optimal in terms of chance of survival).

Even more extraordinary, the AI also proposed a 100% reliable method to avoid dying from breast cancer! It recommended smoking at least five packs of cigarettes per day, whatever the brand. Indeed, data analysis proved this particular cure to be very efficient as heavy-smoking women in the study died from lung cancer before developing breast cancer!

This example clearly shows that Artificial Intelligence lacks contextual analysis, and aims only at solving the problem they have in front of them.

Medical teams are quite reluctant to using Artificial Intelligence to aid diagnosis. The previous example, although obvious, shows that conclusions from machines should always be interpreted within a larger context, because the entire scope cannot be modelled.

Human judgement is certainly key for diagnosis. Yet AI has the ability to digest billions of diagnoses in a few seconds from all over the world, and apply correlation techniques to provide the best possible diagnosis for the patient. Even the worst-case scenario that occurs once every million cases, which would likely not be detected by the average physician, would be taken into account. I am not talking here about entrusting AI to decide medical treatments by itself. However, providing physicians with some recommendations to assist them in making their diagnosis may help. Their expertise will remain the same and medical acts and procedures will remain key. As an illustration, Roche Diagnostic conducted a trial during winter 2018-2019 in Grenoble Hospital to assess the productivity gained by using an automated diagnosis station to diagnose seasonal flu. Their automated diagnosis station allowed emergency staff to regain 30% of their time, using a single station.

IBM’s debater

What distinguishes most humans from machines? Our ability to gather information, structure them in a coherent manner, and have the ability to present them and to debate with others, right?

After having solved the most complex games (jeopardy, chess, etc…), and successfully winning against human players, this is exactly the challenge IBM wanted to face building its “debater”.

The challenge for IBM was to build a machine capable of debating any subject with a professional human debater. The rules were simple: the two opponents had 15 minutes to prepare their arguments about whether “We should subsidize preschools”. AI was supporting the affirmative, Harish Natarajan, its opponent, a widely recognized debate champion, was a grand finalist at the 2016 World Debating Championships and winner of the European Debating Championship in 2012, defending the negative.

The duel appeared to be imbalanced, the machine having immediate access to countless amounts of information, and an infinite computing capacity compared to Natarajan’s notebook and pen from another era.

IBM’s debater structured its four-minute presentation in three major parts and was even able to contradict some of its opponent’s argument. To do so, IBM endowed the system with the following capabilities[2], which is a real achievement; because, a few years back, no AI would have been capable of interacting with a human in real-time with such fluidity:

  • Data-driven speech writing and delivery: ?Project Debater is the first demonstration of a computer that can digest massive corpora, and given a short description of a controversial topic, write a well-structured speech, and deliver it with clarity and purpose, while even incorporating humor where appropriate.
  • Listening comprehension: the ability to identify the key concepts and claims hidden within long continuous spoken language.
  • Four minutes of persuasive speech: the guarantee of producing four minutes of persuasive speech.
  • Modeling human dilemmas:?modeling the world of human controversy and dilemmas in a unique knowledge representation, enabling the system to suggest principled arguments as needed.?

After the debate, the 800 spectators chose Natarajan as the winner. The audience made their choice based on Natarajan’s ability of creating empathy with them, mastering the power of emotions.

The philosopher, Aristotle developed three essential skills more than 2 000 years ago, which constitute the fundamentals of persuasion:

  • The Logos: the ability to argue, and logically structure information,
  • The Pathos:  the ability to create emotions, share them with the audience and create empathy with them.
  • The Ethos, which is the image the speaker projects to his/her audience.

IBM debater clearly did well on the Logos, but still has a lot of progress to make on the Pathos aspect of the debate.

IBM’s project should not aim at replacing humans, but at supporting their natural debating ability.

The replay of the debate between Natarajan and project debater[3] can be found in the references section.

Schubert’s unfinished symphony

Another difference between a human and an artificial brain is the human capacity to create. Some famous composers died before completing their work, and many symphonies were left incomplete. This is the case for the famous unfinished symphony No 8 by Schubert[4].

Schubert only had time to finish the two first movements, and barely started the third one, leaving only 120 measures for piano, with no orchestral arrangements.

There were several attempts to complete the composition of this symphony during the 20th century from various composers, none of them reaching the maestro’s level.

In 2019, Huawei built Artificial Intelligence to create hypothetical melodies for the third and fourth movement. Composer Lucas Cantor rearranged the orchestral composition of the AI.

The symphony was performed live in London in February 2019, and the result was far away from expectations, distant from Schubert’s unique style and sounded more like a commercial blockbuster than an actual musical masterpiece.

Myth or Reality?

Although there has been tremendous progress in getting AI to perform human tasks, they are far from being able to “think”.

They do have access to more knowledge than humans have, are able to structure information and follow a logical plan during presentations, and as IBM demonstrated with their debater project, also has the ability to debate.

But AI are still “learned fools”, hyper-specialized and without memory.

They are still lacking “common sense” and physical and psychological intuitions, our famous “gut feelings”, and still face difficulties with uncertainty, new problems or poorly defined ones, as well as with complex or changing contexts.

Most of all, AI does not “feel”, and is not conscious, and their physicality is still weak (posture, gestures, intonations, …)

AI wont deprive of us the ability to think for ourselves any time soon, even if it becomes more and more useful in assisting us with our most complex tasks. They are a powerful tool, and should not be considered as more than that.

Thanks

I would like to thank warmly Jean-Eric Michallet who supported me writing this series of articles, and Kate Margetts who took time to read and correct the articles.

I would also like to give credit to the following people, who inspired me directly or indirectly:

  • Patrick GROS – INRIA Rhône-Alpes director
  • Bertrand Brauschweig – INRIA AI white book coordinator
  • Patrick Albert – AI Vet’ Bureau at AFIA
  • Julien Mairal – INRIA Grenoble
  • Eric Gaussier – LIG Director & President of MIAI

[1]https://www.lesechos.fr/12/03/2018/lesechos.fr/0301322025116_le-chu-de-strasbourg-s-allie-a-quantmetry-face-au-cancer-du-sein.htm

[2]https://www.research.ibm.com/artificial-intelligence/project-debater/how-it-works/

[3]https://www.youtube.com/watch?v=m3u-1yttrVw

[4]https://en.wikipedia.org/wiki/Symphony_No._8_(Schubert)

By Philippe Wieczorek

Director of R&D and Innovation, Minalogic 

Leave a Reply

Your email address will not be published. Required fields are marked *