Microsoft’s long, winding, and sometimes strange road from keyword to large language model (AI/GPT) chatbots
Having dabbled with AI-ish problems decades ago as a graduate student with simple models and notions coded using (mostly) LISP or Prolog, I found the sudden explosion of end-user accessible AI powered tools in late 2022 fascinating and promising. Reference: Will Microsoft Barney meet ChatGPT? And, this hopefulness was despite the factual errors noted in interactions with OpenGPT.
When I wrote a hopeful notion of adding ChatGPT-like AI to interactive toys like the nearly three decade old Microsoft Barney, I conveniently neglected to note Microsoft’s long and perplexing history of “assistants” stretching back nearly 30 years. Each of the six past assistants ended in failure. And, two (Tay and Zo) took a dark turn before they were terminated. Microsoft’s Bing AI chatbot recent dark turn “prompted me” (pun intended) to take a look back and write this list of Microsoft assistants starting with tiny language models to its present large language model efforts. Reference: MobileViews Podcast 441: Don’t press the academic panic button because of ChatGPT & other end-user AI tools.
If you have been following the trials and tribulations of OpenAI’s OpenGPT, Microsoft’s Bing AI chatbot, and Google’s (as yet unreleased) Bard, the main issue has been incorrect answers to questions. Microsoft’s AI though, is singular in its dark turns. And, this is not the first time a Microsoft chatbot has gone down a strange path. So, here’s a quick overview of Microsoft’s attempts at conversational tools starting with Tiny Language Models to current day Large Language Models.
- 1995–1996: Microsoft Bob. For most users, Bob didn’t really add enough value to be widely adopted. It could be somewhat annoying. And, I recall one reviewer calling it “patronizing”.
- 1997–2000: Microsoft Barney: I reminisced about this toy for pre-readers that I bought for my daughter when she was a toddler. It was, unfortunately, introduced decades before the appropriate technology for it was readily available.
- (Office) 1997–2003: Clippy, Like Bob, had limited usefulness and was annoying. Most people who knew how turned it off soon after installing Microsoft Office.
- 2016: Microsoft Tay was a conversation bot on Twitter that quickly took a dark turn and was turned off after running just one day. As CBS News reported: Microsoft shuts down AI chatbot after it turned into a Nazi
- 2016–2019: Microsoft Zo was a chatbot available via Twitter, Facebook Messenger, and GroupMe. Although it remained available for several years, I don’t know anyone personally who interacted with it aside from a quick test. Like Tay, it spoke out of turn. As Engadget wrote: Microsoft’s “Zo” chatbot picked up some offensive habits
- 2014–2021: Cortana was designed for use on Windows Phone,Windows (desktop), and a short-lived Microsoft smart speaker product. After the failure of Windows Phone (where it might have been useful), it was not an appealing solution for most desktop/laptop users. I don’t recall it taking a dark turn like Tay, Zo, and the currentn Bing AI chatbot. But, then again, this may be because so few people actually used Cortana.
- 2023-???: This brings us to the current Bing AI chabot. And, well, as Engadget wrote: Microsoft’s Bing is an emotionally manipulative liar, and people love it. I will admit that I’m in the camp that finds Bing AI interesting when it is not being evil.
I should note that I haven’t mentioned Apple’s Siri. This is because while it was very interesting prior to Apple buying the technology from the original developers in 2010, it has since remained static in terms of its abilities and knowledge. To be fair, Siri does not claim to be intelligent. When, I asked it “Hey Siri, are you smart?”, its response was “I’m not a person, so I only know what I’ve been progreammed to understand.” Don’t turn evil, Siri!