Ethical Considerations of AI Technologies

As we move into a new computing era, technology has outstripped society in its ability to up-end labor markets and industries, as well as personal lives.

In a panel discussion at Data Summit 2019, Sue Feldman, president, Synthexis, David Bayer, Executive Director, Cognitive Computing Consortium, Tom Wilde, CEO, Indico, and Steven Cohen, COO, co-founder, Basis Technology, explored the ethical and legal issues that computing advances have raised. 

Areas covered included security, privacy, decision making by humans vs. machines, and self-driving vehicles.

Discussing the question of whether it is possible to teach fairness, ethics, and human judgment to AI systems, the panel and attendees discussed whether this is even desirable.

Ultimately, it all relates to training data and algorithms put into a system, attendees and panelists agreed, but it may be impossible for people feeding data into systems to be aware of all biases, and if they are not aware of it, they can’t fix it.

It was pointed out that increasingly there is a tendency to trust a machine more than people, and there is an implicit belief that the machine is more likely to be correct, which can make their results even more impactful.

In terms of self-driving vehicles and autonomous machines, panelists and attendees considered how to rationalize decisions when an accident is unavoidable and the possible implications for liability when no human is involved. Decisions will be rooted deep in training data and algorithms making it difficult to determine liability and blame.

Humans also often may make a certain decision in one scenario but another decision on another day—for example crossing the street in a certain way when alone but not doing so when holding a grandchild’s hand.

There is also a paradox of accuracy, the panel noted. A machine can be highly accurate but if it makes a mistake, it makes headlines. If a machine is not 99.999% accurate, it is considered unreliable. However, we don’t hold people to the same standard and if someone is most often accurate and correct, we think they are doing a good job.

Being too good can also be a problem for smarter systems, the panel noted, referencing the famous Target marketing campaign which shocked a family when it sent a pregnant teenager its offers.

Ultimately, it may be necessary for government to step in with regulations as far as what is acceptable with advertising campaigns.

Many presenters are making their slide decks available on the Data Summit 2019 website at