AI and ethics goes back to my youth and reading science fiction stories such as I, Robot and The Three Laws of Robotics by Isaac Asimov and later in my teens watching movies like Wargames, RoboCop and Terminator where machines either learned about ethics or had none, just a primary objective.
But today the conversation about ethics and AI is much more complex than just killer machines and touches all of us in our personal and professional lives. I have highlighted four areas which currently find of interest and will hopefully bring some further insights to the subject:
Bias – Can machines avoid bias?
There is no single reason why AI could have a bias when providing answers or outcomes.
It may be due to insufficient training data where certain demographic groups are either missed completely or under-represented. This can lead to AI not recognising or ignoring certain groups which are not represented and could have serious consequences if scanning medical information.
However training data is not the only cause of bias and humans can also carry a level of (un)conscious bias which can affect AI. These biases can find themselves in historical training data when the system is being trained and then further reinforced by the result. A recent report in The Guardian reported how images of women on social media were much more likely to be highlighted as suggestive and suppressed.
To mitigate bias developers need to ensure that data sets and training are diverse and auditing is done to catch any bias which might be appearing. If done right this could be of huge benefit to society as removing bias can only be a positive.
Privacy – What information can AI collect?
AI is all around us and can be used in multiple ways to improve our lives. We use voice recognition to control our houses and cars but who is controlling what is being collected, either intentionally or unintentionally, by a device? Facial recognition is another area where privacy is an emerging issue. Used at a large scale with little or no consent raises all types of issues about the data collected. Would attending a protest count poorly towards a social credit score? Alternatively it could be seen as a huge positive in identifying and preventing known terrorists or criminals committing an act.
To counter the ethical issues of how this type of data is collected and stored many governments have decided to enact temporary bans on facial recognition. For example, in 2020 the European Commission banned facial recognition technology in public spaces for up to five years to make changes to their legal framework.
Control – Who should be in control of AI?
AI has been used by the military for many years in the delivery of weapon systems to destroy targets but of course now is being used more in other systems such as the drone for package delivery. But there has always a need for human control and moral decision making according to some set of rules.
The concern that we now see now is AI is increasingly used in decision making at faster speeds, often when humans may not have the physical reaction time. This might be a financial services system where rapid decisions are needed based on massive amounts of trading data. More controversial is when considering autonomous cars and how to make a decision of who could be injured in an accident. This Forbes article provides some good insights into this question but I firmly believe that it must still be left to a human to decide. This leads nicely to my last point on accountability.
Accountability & Ownership – Who is responsible for AI decision making?
Who is accountable for the outcome of AI results? Is it the programmer, the trainer, the decision maker, the vendor? The list could go on but the simple answer at the moment is not clear.
What is required for companies that do deploy and use AI technologies is to have an appropriate governance framework around their use. This should cover topics such as the above of bias, privacy and control. Organizations that are aware of this are far more likely to identify issues early in systems and avoid unpleasant surprises.
Finally, of course, what role will legislation play in this and help drive a transparent framework? The EU is driving the charge as it did with the processing of personal data and is using a similar approach to segregate AI into different levels of risks and associated controls required.
All the above plus many more areas are relevant as to the discussion of how AI will be used, maintained and legislated, and I could not finish without a quote from the most topical AI engine at the moment – ChatGPT. I asked ChatGPT “Does ChatGPT have any ethics?” and this was the response:
As an AI model developed by OpenAI, ChatGPT does not have personal beliefs, emotions, or ethics. It is an artificial intelligence system designed to process and generate text based on patterns it learned from the data it was trained on.
However, the ethical considerations associated with the development and use of AI systems, including language models like ChatGPT, are an important topic for discussion. OpenAI and other organizations have issued guidelines and principles for the responsible development and use of AI, which can help ensure that these technologies are used in ways that are safe, fair, and aligned with human values.
It’s worth noting that the training data used to develop language models like ChatGPT can have a significant impact on the model’s output, and it’s important to ensure that the data is diverse and free of harmful biases. Additionally, the use of language models in sensitive applications, such as generating fake news or biased content, can have negative consequences and should be approached with caution.
Iain Cunningham has over 20 years working in digital transformation and how technology can change the way we work. He is particularly interest in robotic process automation and machine learning and enjoys reading about ethics and AI.
