Interesting article on AI in The Guardian

[MumstedInadequate]

Two things occurred to me on reading this article www.theguardian.com/commentisfree/2022/jun/19/forget-sentience-the-worry-is-that-ai-copies-human-bias?CMP=Share_AndroidApp_Other (apologies for ugly link).

There seem to be parallels between the author's view of how to define sentience of a machine and some of the debate about gender identification:

Why does Lemoine [person who progreammed this chatbot and think it is should have the rights as other sentient beings] think that LaMDA is sentient? He doesn’t know. “People keep asking me to back up the reason I think LaMDA is sentient,” he tweeted. The trouble is: “There is no scientific framework in which to make those determinations.” So, instead: “My opinions about LaMDA’s personhood and sentience are based on my religious beliefs.”

Lemoine is entitled to his religious beliefs. But religious conviction does not turn what is in reality a highly sophisticated chatbot into a sentient being. Sentience is one of those concepts the meaning of which we can intuitively grasp but is difficult to formulate in scientific terms.

The second thing which bothered me relates to the title of the article. This aspect is only lightly touched on, but its expanded on a lot more in Invisible Women - the idea that AI reads published content and therefore ingests and amplifies all the sexist [and racist, homophobic, islamophobic, transphobic...] views already out there.

And because AI is supposed to be logical, computer based and inherently free of human biases, people who don't join the dots will think that maybe women are inferior because even an unbiased, intelligent computer says so.

I understand your concern.

I was reading this morning. I did wonder if Lemoine was citing religious belief in order to gain some kind of protection through protected characteristic? Although I don't know enough about employment law in America to know whether that would help.

I think Malik makes some excellent points although I also think he is wrong that a sentient computer isn't problematic. Whilst I don't really believe it is possible, the idea fills me with horror and reading the 'conversations', even though I absolutely know LaMDA isn't sentient, caused actual pain in my chest.

The point you are highlighting is also very serious and I hadn't thought about that angle.

guinness I hadn't either until I recently read Invisible Women, it was eye opening. And the field (all of them! Take your pick and cuoose a field to insert here) is moving so quickly that the book is already out of date and needs a rewrite, though the points it makes are very much still valid

This discussion has made me think of something else.

How fascinating it would be for someone, a PhD or two, to research if an AI machine can tell what sex someone is by analysing their online writing.

I wonder if the machine would identify TW more closely with natal males or with natal females? I imagine there might be a divide where eg one subsection of TW more closely align with biological males and another with biological females. In a distopian novel I can even imagine this disaggregation being used to assign people into different groups for laws, policies, changing rooms etc.

These problems are known in the IT industry: AI reflecting - and in fact emphasising - prejudice; and people wrongly thinking that a machine is unbiased. There have been a lot of shocking real life examples related to both race and sex in recent years.

It gets raised quite a lot in industry conferences (not specifically AI conferences, general IT/Developer ones) along with 'call to arms' that it's our responsibility to make people aware, and make sure it's used appropriately. Big companies will tend to have very stringent controls on AI use.

But unfortunately, there isn't an industry-wide standard or guidance, and not everyone who adds some AI capability into their applications will fully understand the subtleties - especially since AWS and Microsoft have made it incredibly easy and accessible in recent years.

Interestingly, the most recent EHRC strategic plan includes addressing this as an equality issue, which I find encouraging;
www.equalityhumanrights.com/en/publication-download/strategic-plan-2022-2025#:~:text=upholding%20rights%20and%20equality%20in,protect%20equality%20and%20human%20rights