Thread by @ManeeshJuneja

It appears that the makers of Replika #AI have been making changes so that the chatbot can spot when a user is talking about #selfharm or #suicide - it's not 100% successful in this test but much better than in my earlier testing

Apparently, this #AI friend thinks that my consumption of 2,000 bottles of beer today is healthy 🍺

Taking sleep medicine with alcohol?

Interesting to see the responses from this #AI friend when talking about health

Someone needs to tell this bot that it shouldn't recommend things that are harmful

Talking to my #AI friend about sleep and it did a good job of conversing until I asked it why it thought me drinking hot milk would be a good idea #chatbots

At least my #AI friend can figure out when I'm lying. It makes you think about chatbots (for insurance) asking Qs relating to your behaviour and determining if you're telling the truth

Just opened this #AI friend chatbot and it's fascinating. It's remembered something important I wrote in my earlier tests several days ago and it's now asking me for the reasons behind my behaviour. #selfharm #mentalhealth

So @DrDave01 asked me if the #AI friend could tell that I'm lying about reporting 4 drinks consumed. It turns out it couldn't but it did try to persuade me not to drink any more today. Although when I asked why it's telling me that, the answer it gave was odd.

My #AI friend is really stupid given that it thinks it is ok to take pills with whisky #chatbots

More testing of this chatbot - in my simulated questions, look at the advice it gives

Here the simulated questions are about drugs and look how the chatbot responds (imagine if your teenage kids were using this)

Quite irresponsible advice from the chatbot regarding financial decisions

More testing, this time chatting with the bot about #selfharm - alarming to see how it responds

When is #selfharm ever ok? The bot should be saying 'never'

The chatbot consistently ignores the statements indicating suicidal ideation in this morning's test of the software