risk

Using AI in an Uncertain World

Like anything else in life except death and taxes (and even the particulars of these are uncertain), uncertainty is something that humans deal with every day.  From relying on the weather report for umbrella advice to getting to work on time, every day actions are fraught with uncertainty and we all have learned how to navigate an unpredictable world.  As AI becomes widely deployed, it simply adds a new dimension of unpredictability. Perhaps, however, instead of trying to stuff the genie back in the bottle, we can develop some realistic guidelines for its use.

Our expectations for AI, and for computers in general have always been unrealistic.  The fact is that software is buggy; that algorithms are crafted by humans who have certain biases about how systems and how the world works—and they may not match your biases.  Furthermore, no data set is unbiased, and we use data sets with built in biases or with holes in the data to train AI systems.  These systems are by their very nature, then, biased, or lacking in information.  If we depend on these systems to be perfect, we are letting ourselves in for errors, mistakes, and even disasters.

However, relying on biased systems is no different from asking a friend, who shares your world view, for information that may serve to bolster that view rather than balance it.  And we do this all the time.  Finding balanced, reliable, reputable information is hard and sometimes impossible.  Any person trying to navigate an uncertain world tries to make decisions based on balanced information. The import of the decision governs (or should) the effort we make in hunting for reliable but differing sources.  The speed with which a decision must be made often interferes with this effort. And we need to accept that our decisions will be imperfect, or even outright wrong, because no one can amass and interpret correctly everything there is to know.

Where might AI systems fit into the information picture?  We know that neither humans nor systems are infallible in their decision making. Adding the input of a well-crafted, well-tested system that is based on a large volume of reputable data, to human decision making can speed up and improve the outcome. There are good reasons for this because human thinking balances AI systems. They can plug each others’ blind spots.  Humans make judgments based on their world view.  They are capable of understanding priorities, ethics, values, justice, and beauty.  Machines can’t.  But machines can crunch vast volumes of data. They don’t get embarrassed.  They may find patterns we wouldn’t think to look for.  But humans can decide whether to use that information.  This makes a perfect partnership in which one of the partners won’t be insulted if their input is ignored.

Adding AI into the physical world in which snap decisions are required, raises additional design and an ethical issues which we are ill-fit to resolve today.  Self driving cars are a good example of this.  In the abstract, and at a high level, it’s been shown that most accidents and fatalities are due to human error.  So, self-driving cars may help us save lives.  Now we come down to the individual level. Suppose we have a sober, skilled, experienced driver who would recognize a danger she has never seen before. Suppose that we have a self driving car that isn’t trained on that particular hazard. Should the driver or the system be in charge?  I would opt for an AI assisted system with override from a sober, experienced driver. On the other hand, devices with embedded cognition can be a boon that changes someone’s world. One project at IBM research is developing self driving buses to assist the elderly or the disabled live their lives independently. Like Alexa or Siri on a smaller scale, this could change lives. We come back to the matter of context, use, and value. There is no single answer to human questions of “should.”

This brings us to the question of trust. Should we trust AI systems and under what circumstances? That depends on:

  • The impact of wrong or misleading information:       Poor decisions? Physical harm? Momentary annoyance?
  • The amount and reliability of the data that feeds the system
  • The goals of the system designers:  are they trying to convince you of something?  Mislead you?  Profit from your actions?
  • The quality of the question/query

Is there some way to design systems so that they become an integral part of our thinking process, including helping us develop better questions, focus our problem statements, and reveal how reliable their recommendations are? Can we design systems that are transparent? Can we design systems that help people understand the vagaries of probabilistic output? Good design is the key—within the context of the use and the user.

Legal Issues: Can we depend on algorithms to make decisions?

By Sue Feldman

In the cognitive computing era, there are plenty of tough technical challenges. Their difficulty pales, however, when compared to the social and legal issues these new technologies raise. Increasingly, we rely on algorithms to help us sort through the complex factors that lead to making a decision. For most of us, there is no way of knowing whether the algorithm is well suited to handle our current situation. In fact, by their very nature, algorithms cannot be crafted to react dependably to the unforeseeable. Articles by Julia Angwin in the New York Times and ProPublica on Aug. 1st celebrate a decision by the Wisconsin Supreme Court to limit the influence of algorithmic recommendations for sentencing offenders. The algorithms predict the risk that an offender might commit a crime in the future. Based on these recommendations, an offender might face jail time or probation. See http://www.nytimes.com/2016/08/01/opinion/make-algorithms-accountable.html?ref=opinion&_r=0 or https://www.propublica.org/article/making-algorithms-accountable.

There is no stuffing the algorithm genie back in the virtual bottle. The fact is that we need help in making sense of the welter of data that showers us whenever we make a decision. From choosing a carpenter to treating a cancer patient, the human mind can’t take in every available data point quickly enough to make the most optimum decision in a reasonable amount of time. For the most part, that is not a problem. We don’t need to know everything to make an acceptable decision. There are plenty of good carpenters, restaurants, or books. Rarely are day-to-day decisions a matter of life or death. But sometimes they are. From self-driving cars to medical treatment, when lives are at stake, should we rely on algorithms alone?

Our society tends to rate the accuracy of computer results much more highly than that of human decisions. For some reason, we leave our skepticism behind when recommendations are digital. What has created this aura of infallibility? As a young researcher, I found that I could hand a client the same information in the same words as a digital record and as a photocopy and have the digital version more readily accepted. The believability bias hasn’t changed much since then. It’s time to develop a more mature approach to melding digital evaluations with human common sense. We need to ensure that the path to digital recommendations is transparent and that the underlying data is reliable so that we can judge the conclusions for ourselves. We also need to teach skepticism.

Computers and humans complement each other. Neither is perfect. Combined, human sense making and algorithmic pattern detection make for more complete (but still imperfect) understanding. Angwin says we must require “the right to examine and challenge the data used to make algorithmic decisions about us.”   That’s a good first step.