Ethics of AI

There is no bad AI

The two different kinds of AI, narrow and general, raise very different ethical considerations. There has been a lot of media hype about the dire ethical implications of general AI in particular, including existential threats to humanity from the technology. Beyond the hype there has been a great deal of work done by thought leaders in the field of AI. Organizations such as the Future of Life Institute, Machine Intelligence Research Institute and Partnership for AI, are just a few of the groups that have been looking into the ethical issues related to AI and have defined principles to guide future work in the field. Much of the impetus for this forward thinking has come from lessons learned by the physicist who developed the technology underlying nuclear weapons and would later regret not sufficiently considering the ethical implications of its awesome power.

While I applaud these efforts to think ahead, the development of general AI is likely at least 20 years away, if not more. On the other hand, narrow AI is already being widely used today, and there are many real ethical issues right in front of us related to this technology. So this article will focus on the ethics of narrow AI.

Plan for failure

First, let’s consider unintended bad consequences associated with AI. Usually, ethical responsibility requires some knowledge or foreseeability of a potential harm. The ethical responsibility to consider the potential harm of a technology clearly lies with its developers. Hiding behind ignorance is not a shield to this ethical responsibility. This means developers must actively seek out an understanding of the possible harms from AI systems using simulations and other techniques. Assuming this level of responsibility for foreseeable risks is a higher bar than many tech companies have taken in the past.

Some AI failures are expected but many are not. Corner cases by definition are not planned for. Nonetheless, we know that AI will fail, we just may not know how or why. An ethical system must plan for failure. This is especially important when considering the impact of failure, such as self-driving cars which may cause injury or death when they fail. Being reckless of risks is clearly not ethical. If there are risks that cannot be mitigated, informed consent of the user or others that may be adversely affected is essential. [pause] In addition, as we saw with predictive policing applications, it is the ethical responsibility of the creator of those AI systems to correct any improper bias in training data to ensure an equitable result.

AI is a powerful tool for bad actors

Putting aside the unintended consequences of AI, let's talk about when AI is used intentionally is a way that may be unethical. These are not situations from science fiction, when a general AI becomes conscious and causes havoc. In these situations, the AI is not bad... nor good for that matter. It is the person or organization wielding the AI that is bad. These are situations where an effective AI is being used by someone in an unethical way. That bad actor may be a government, a company, or an individual.

You should be able to hide your face from the government

The first category of potential bad actors is when governments use AI for their purposes, even if legitimate, in ways that may have ethical implications. One government purpose is the surveillance of its citizens to defend against crime and terrorism.

China, it seems, is testing the outer boundaries of how governments can ethically use AI in terms of surveillance of its own people. China is building the world’s most powerful facial recognition system with the objective of identifying each of its 1.3 billion citizens within three seconds, and matching a person’s face to their ID photo with at least 90% accuracy. The system feeds off a network of over 170 million cameras installed around the country, with an expected 400 million more coming online. Chinese start-ups are being complicit in this effort, and have built algorithms that the government uses to track members of a largely Muslim minority group in the western part of the country.

Meanwhile, courts in the US have ruled that this level of surveillance is a violation of fundamental privacy rights. Nine cities in the US have banned the practice for their police. The notion is that you can leave your mobile phone at home, but you can’t leave your face.

AI technology has fundamentally shifted the balance of power between individual and government. On this balancing decision, human rights experts, policy makers and philosophers, and pretty much everyone needs to weigh in. This is why it's very important for individual citizens to be well schooled on how AI works, what it is capable of doing and what its limitations are. The best defense against ethical violations here is a well-informed population awoken to the power of AI technology. Ultimately, all governments must be accountable to their people.

Marketing using AI may cross the line into manipulation

The second category of bad actors is when a company uses AI in a manner that may not be ethical. Commercial use of AI may end up violating expectations of privacy when data is combined and used in ways they were not expected. AI should use data provided by consumers or businesses only with their informed consent.

One of the more ethically challenging areas of commercial AI applications is consumer marketing. Traditionally, marketing was a very imprecise science, and this slippage allowed users to maintain their privacy and separation from the marketer. At the same time, much of the message was unhelpful and the cost of misdirected messaging was wasted. This will change as AI-powered marketing completely closes the loop and precisely targets individuals in one-to-one marketing. This kind of marketing changes the relative power between individuals and marketers. It starts to seem that consumers are outgunned and the arrangement unfair. Humans won't have the ability to resist marketing messages or even know what's happening as their emotions and subconscious are gamed by a super-attentive AI. Cambridge Anaytica using Facebook user data to inflame people’s emotions and political views was a crude example of this [pause] The good news here is that consumers are increasingly more awake to this issue and ultimately we can regulate the activities of these companies.

AI is a weapon of mass destruction for the lone wolf

The final category of AI ethical concern may be the most dangerous. This is when AI is used by individual bad actors. These actors can, acting alone, use AI to inflict disproportionate harm. Governments and companies can be held accountable ultimately, but these people are hard to find and even harder to hold accountable. Unlike nuclear weapons or other weapons of mass destruction, the scale necessary to create an AI is very small. One individual using off-the-shelf technology and readily available public data could cause a lot of harm.

We see this in DeepFakes, videos created by AI, made to look real but with swapped faces and words. DeepFakes spread around social media may create a cloud of hostility, misinformation and confusion, and lead to outrage or even panic. In addition, AI may be used to attack key systems as they become more linked online.

The best defense against these attackers is a strong offense. We will need to develop bots to fight the bad actors’ bots. This is already the case in cyber security and fraud detection. In this war of bots, there is no guarantee that the good bots will win in every case, and they will likely fail on occasion.

Get smart and be vigilant

Our society will need to become broadly equipped and educated to handle all of these AI-related ethical challenges. AI itself may even prove to be a powerful ally in this effort. Ultimately, I believe the benefits of AI will far outweigh the burdens, but we will need to be vigilant to ensure that the benefits are shared broadly, and we do not lose our humanity, our security or our freedoms in the process.