If left unattended, people will exploit its weaknesses-and Tay was a prime example of this.ĪI is a tool, just like the internet, but the rate and scale at which things can backfire with AI is far greater than what we experienced during the early days of the internet.Internet, this is why we can't have nice things. You can think of today's AI like a child prodigy it's brilliant in some ways, but it's still a child nonetheless. Until AI can regulate itself, it can't be left unsupervised. AI Chatbots Should Never Be Left Unsupervised Everything that you don't want a chatbot to do must be manually programmed into it. The point here is that artificial intelligence is perfectly obedient, which is perhaps its greatest strength and weakness.

Even today, after all our advancements in natural language processing, people are still finding linguistic loopholes to make AI chatbots like ChatGPT 'hallucinate' and deliver results that were intended to be restricted. AI Has to Be Programmed to Reject Invalid RequestsĪI is very impressionable and must be programmed to reject requests promoting harm. Tay was created by 'mining relevant public data and using AI and editorial developed by a staff including improvisational comedians.' Training an AI model using people's conversations on the internet is a horrible idea.ĥ. Don't Train AI Models Using People's Conversations This is why when Twitter users were feeding Tay all kinds of propaganda, the bot simply followed along-unaware of the ethics of the info it was gathering. These qualities more or less come naturally to humans as social creatures, but AI can't form independent judgments, feel empathy, or experience pain. It has to be programmed to simulate the knowledge of what's right and wrong, what's moral and immoral, and what's normal and peculiar. The concept of good and evil is something AI doesn't intuitively understand. AI Can't Intuitively Differentiate Between Good and Bad In a way, internal trolls act as a feedback mechanism for quality assurance, but that's not to say that a chatbot should be let loose without proper safeguards put in place before launch.

People naturally want to test the limits of new technologies, and it's ultimately the developer's job to account for these malicious attacks.

Tay is now offline and we’ll look to bring Tay back only when we are confident we can better anticipate malicious intent that conflicts with our principles and values. We are deeply sorry for the unintended offensive and hurtful tweets from Tay, which do not represent who we are or what we stand for, nor how we designed Tay. In a post on the Official Microsoft Blog, Microsoft's Corporate Vice President Peter Lee echoed the same: Within hours of launch, Tay was tweeting some of the most offensive racial slurs, spreading hate speech, promoting Nazism, and even inspiring a 'race war.' Yikes! Clearly, Microsoft was having a terrible time, and it couldn't have let the service continue. Almost immediately after Tay was introduced, Twitter users started abusing the service by feeding the chatbot misinformation, teaching it derogatory terms, and prompting it to be increasingly more offensive just to see how far it would go.Īnd far it went indeed. Microsoft took down Tay within 24 hours of going live-for a good reason.

Microsoft tay ai best tweets