Bad AI Chatbot Advice to Recovering Alcoholic

Google’s health AI was advising an alcoholic to drink a beer an hour after a relapse - and constantly messaging them with more bad advice

By Nick Jonsson

As a sobriety coach, psychotherapist and counselor and someone who has struggled with alcohol addiction in the past, this recent story really stopped me in the tracks.

A user made a post on Reddit within a community for recovering alcoholics called stopdrinking - about how their AI chatbot, following a relapse, began throwing out a range of really bad, very dangerous advice. 

The AI chatbot which was apparently powered by Google was advising the person to “drink a beer an hour” and messaging them constantly to check that their fridge was well stocked with beers, warning them that they risk dying - and clearly putting the person on edge.

The post was screenshot and then shared over on social media platform Bluesky where many thousands of people reacted with anger, mainly towards the AI bot, but sadly also some was aimed at the person for trusting an AI chatbot in the first place.

The truth about relapse - and severe, sudden alcohol withdrawal

Before it looks like I am defending AI here (which I’m not trying to do), there is a common misunderstanding around withdrawal for an alcoholic - and that it is safe for them to go cold turkey. 

If someone has been a long-term heavy drinker, there is genuinely a risk of them getting what is known as delirium tremens, which is a severe, life-threatening medical emergency which can be caused when this person suddenly stops drinking alcohol. 

And so this is likely the reason (or the misunderstanding) that the AI chatbot has had, when the user told them about them giving up alcohol. The suggestion of drinking a beer an hour is clearly bad advice, as is telling them to stock up their fridge with beers, but it is possible for us to see where this misunderstanding comes from. 

The dangers of using an AI chatbot for supporting their recovery from alcohol

I have spoken out before about the risks that AI chatbots have when being used to treat serious health issues for users - I feel so strongly about this that I created NickGPT - my “human” take on an AI chatbot. 

The hope was to appeal to the types of people who want to have the ease of use of a tool like an AI chatbot - but to actually get one-to-one, real, human advice, whilst referencing many instances of AI giving dangerous advice.

And having struggled with alcohol addiction myself almost a decade ago, I know how vulnerable someone can be when they reach out for help. My recovery happened before AI became part of everyday life, and it wasn't technology that changed my life. It was walking into a room full of people who had been exactly where I was. It was that human connection, the understanding, honesty and accountability that helped me recover.

When I finally decided to stop drinking, I couldn't simply quit by myself. I required medical treatment and was admitted to hospital for a medically supervised detox. That wasn't a sign of weakness – it was the safest and most appropriate way to begin recovery.

In my opinion, the only valid way an AI chatbot could be used in such a situation, is for it to recommend that the user reaches out to a health professional for a proper diagnosis and to take appropriate action - not for the chatbot itself to begin diagnosing and giving out advice like it appears to have happened here.

Reaction to how people responded to this incident online

Reading comments on the Bluesky post where this gained a lot of traction (and I note the original Reddit post had been deleted by a moderator, although it’s not clear why) - I’m quite disappointed by how many people have responded online.

A selection of some of the comments found over on Reddit

Lots of people have criticised the person for going to an AI chatbot and seeking advice. For me this is the wrong thing to criticise - this is a very vulnerable person, who is probably going through a very difficult period in their lives, who probably just took the easy option - to reach for his phone and to start asking these questions. 

I think it’s wrong of us to look down or make fun of this person who has probably just fallen for a lot of the hype around AI, which is easy to do when so many big corporations and industries have been so quick to adopt it and promote it. 

Instead I think we should credit this person for being brave enough to try and tackle their drinking problem, for finding a community of (hopefully) real people on Reddit who seem to be supporting and helping each other, and for also sharing their story as a way of warning people about their experience using AI to stop drinking.

A quick look on this Reddit group shows that this person is not the only one who has been talking about engaging with their AI chatbot to get help with their recovery from alcohol - and so sadly, there may be many other vulnerable people being misdiagnosed or being given very dangerous medical advice.

Thankfully many of these posts are about people using ChatGPT more of as a way to journal and process their journeys away from alcohol, not to get medical advice or therapeutic support, and they do mention that they are doing this in addition to going to in person local events to talk face-to-face with others - so this is on top of what they are currently doing.  

Getting real sober coaching with me - not an AI chatbot

As a reminder, I do offer sober coaching for people looking to take control, rebuild energy and align your life with purpose. I’ve been here before and have made it through to the other side - if you want to speak to me about this, just reach out!