Putting AI Into a Black-box of Super-Ethics

Imagine, in the near future, a GP using a machine learning algorithm to recommend a patient’s treatment based on their habits, diet, age, genetic history, etc. What happens if the recommended treatment fails? If the machine learning algorithm is based on a complicated neural network then it may prove to be impossible to work out why the algorithm was unsuccessful. However, if it’s a machine learner based on decision trees or Bayesian networks is much more transparent to programmer inspection (Hastie, Tibshirani, and Friedman 2001).

We’re not quite there yet. However, we are making headway to create artificial intelligence (AI) that aids workstreams.

One example is PlasmoTron, go to http://bit.ly/plasmotron to see it in action, a computer program that controls an existing open-source robotic pipetting robot to automate the growth of malaria parasites.

PlasmoTron — automatic pipetting for malaria parasite research by Theo Sanderson, Post-Doctoral Research Associate, Wellcome Trust Sanger Institute.

Keeping malaria parasites alive can take a lot of time. It doesn’t require much thought, but highly trained scientists spend a lot of time just moving liquid from one tube to another to keep their parasites alive. This places limits on how many different parasite lines one can work with at the same time, which limits how many genes we can understand.
— Theo Sanderson, Post-Doctoral Research Associate, Wellcome Trust Sanger Institute.

According to some definitions of AI, PlasmoTron is intelligent — it makes judgments. Still, at a fundamental level, it is just following a simple set of rules to determine how to take care of the parasites in its charge. The process is currently semi-automated, the researcher still has to take plates in and out of the incubator.

Taking the effort out of mundane tasks is undoubtedly the aim of the PlasmoTron and a lot of AI currently out there. This use of AI and robotics is undoubtedly a benefit. There are general misconceptions about robotics often conveyed through popular fictional media such as Blade Runner and Terminator films.

An AI GP dispensing medical advice and medicine proposes moral ethics for society. How is it controlled? What happens if errors occur? One theory is that AI is continuously monitored and tested with more rigorous ethics learned from current practices. Can we also apply ‘black box thinking’ to AI and robots? The black box dramatically reduced aviation deaths from 25% in 1912 to a rate of one accident per every 2.4 million flights (Syed 2016). A feature is becoming popular in motoring. With dashcams and in-car computers, the SEAT Leon Cristobal concept takes this even further potentially incorporating a black box safety recording data and images while driving then sending them to your smartphone if an accident occurs (SEAT 2017).

The prospect of AIs with unimaginable intelligence and abilities offers humanity with the unusual challenge of stating an algorithm that outputs superethical behavior. These problems may seem visionary, but it seems predictable that we will encounter them; they are not devoid of present-day research directions.


Acknowledgments

With thanks to Theo Sanderson, Post-Doctoral Research Associate at Wellcome Trust Sanger Institute for answering my many questions about PlasmoTron.

Eleonora Aquilini for the suggestion, advice, and referral to Theo.


References

Hastie, Trevor, Robert Tibshirani, and Jerome Friedman. 2001. e Elements of Statistical Learning: Data Mining, Inference, and Prediction. 1st ed. Springer Series in Statistics. New York: Springer.

Syed, M., 2016. A Joosr Guide to… Black Box Thinking by Matthew Syed: Why Most People Never Learn from Their Mistakes — But Some Do. 1st ed. UK: Joosr.

Opentrons. 2017. Opentrons. [online] Available at: http://opentrons.com.[Accessed 13 December 2017].

Twitter @OpenTrons_. 2017. @OpenTrons_. [online] Available at: https://twitter.com/OpenTrons_/with_replies. [Accessed 13 December 2017].

Bayesian networks. (2018). Introduction to Bayesian networks. [online] Bayesserver.com. Available at: https://www.bayesserver.com/docs/introduction/bayesian-networks [Accessed 3 Jan. 2018].

SEAT. 2017. Innovative Solutions for Smart Cities. [online] Available at: http://www.seat.com/corporate/news/events/smart-city-expo-2017-leon-cristobal.html. [Accessed 3 January 2018].

Previous
Previous

What are semantic colours?