As the power of AI becomes ever more visible, there are inevitably worries about its use.
This article in C&EN magazine makes a good job of highlighting concerns and their counterarguments. Citrine as a company finds this topic important and would like to contribute to the discussion.
Not all AI is made equal
The level of ethical scrutiny needed should be proportionate to the potential to cause harm to society. We should put great effort into making sure that data used to train AI is not perpetuating historical inequalities. We should ensure we don’t create systems that could create new chemical or biological weapons autonomously. But systems that accelerate scientific discovery by assisting a researcher in choosing the next experiment to carry out, need fewer guard rails in place.
There is a difference between AI systems that generate data or text or images and present them as fact and those that analyze data and make suggestions with explicit probabilities of accuracy.
The role of the user
Have you considered the ethics of Excel? Excel enables the user to make thousands of calculations at the click of a button, in the same way that AI enables a researcher to analyze the statistical probability of thousands of scenarios. You might be thinking that Excel cannot be dangerous… you’d be wrong. Excel is used universally by influential people to create and analyze data that affects all our lives. Here is a great example of a simple excel mistake that had an impact on millions of lives. The fault here though was not with Excel, it was with the users. They didn’t double check their use of formulas. Sense-checking and proofreading your work, whether you are using AI or Excel is important, as lawyers in New York recently found out.
So how do you make sure to use AI responsibly?
- Be open about using AI. Don’t present AI generated content as your own. Be explicit about the source of data.
- Use AI systems that provide tools for sanity-checking such as feature importance statistics that help you see what inputs are influencing outputs.
- Use AI systems that tell you the predicted accuracy of results and never take what AI tells you at face value without engaging your brain.
- Assess potential negative impacts of your work and guard against them. Make sure the data you are using is unbiased. Make sure there is an air gap between calculations and potential negative actions.