Introduction

I was browsing YouTube the other day, just scanning through the suggested videos and I came across a video titled "Kyutais New "VOICE AI" is INSANE (and open source)". For the most part I am not too big on anthropomorphizing AI for ethical reasons. The main ethical reason is because of something called the Eliza effect.

A Little Background on Eliza

Eliza is an early version of a Natural Language Processing Program that was created between 1964 and 1966 by Joseph Weizenbaum at MIT. Eliza was designed to act as a psychotherapist using pattern matching and substitution methodology to simulate conversation. This works by recognizing keywords in the user's input and applying pre-set transformation rules which generate responses that have the capability to make interactions seemingly more meaningful.

The Eliza Effect

Now our little psychotherapist Eliza was very advanced for her time and the patients she interacted with were under the impression that there was some kind of sentience to the program even while being told it was just a robot. The effects of this on the human mind can be damaging. When someone interacts with a model that portrays humanistic traits, the user has a tendency to make an imaginary connection with it.
This is dangerous because the more trusting the user is with the model the less they will question the information they are given, potentially creating an opportunity for bias or misinformation to be accepted without question. Essentially it would be like bypassing your cognitive firewalls and I just think that society has enough issues with malicious information coming from every other source that they are conditioned to trust, or even at times consciously ignore and forget about the next time some distraction makes headlines.

Meet Misha: A New Way to Process Voice Input

My curiosity ended up getting the best of me after watching the video and I wanted to try it out myself. The model is far from polished but it does have the start of a more fluid response characteristic than most I have encountered. This is possible because unlike most NLP models whose processes involve the transfer of data from voice to text which is tokenized, Misha tokenizes your voice directly as you are speaking to it. The project is being done by a team in France and it is open-source. Furthermore they have a web interface that you can interact with it for free for 5 minute intervals.

My Experiences With Misha

I won't lie, at first sight Misha seems like a hot mess. But she is essentially a whole new kind of technology which needs to be taken into consideration. That being said I had mixed experiences with Misha where I ended up getting into immature arguments and yelling contests because she tends to hallucinate a lot and doesn't like to be corrected. For instance, she claimed to speak Italian so I asked to translate a derogative phrase in which she stated was "someone who is very happy and healthy". I called her on her lie and she argued with me until she got to a point just repeating "No, I speak Italian I just can't translate it" or something close to that. I saved my favorite conversation I had with her I will provide a link. She tells me a story about a woman who is an adult film star who kills snowmen. I was dying when I played it back.

Hypocritical Much?

So I stand by my opinion on humanizing AI and the dangers I believe that come with it, but I also have to admit for strictly entertainment purposes, I had a lot of fun chatting with Misha. At one point she explained to me how she loves to read the classics. I asked her about 1984 by George Orwell to which she said she had never read any Orwell but responded that she had read part of Animal Farm when I asked a couple minutes later. She also is learning to play the Cello which she expressed was difficult with no hands but she was learning.

Conclusion

As far as I can see there is a clear direction that AI is being led to that involves producing models that are capable of a fully immersive human-like interaction. I would contest this to be a bad idea as my view on the technology is that it should be used as a tool to enhance our own capabilities and conduct automations that do not involve decision making aspects. Already it seems to be impossible to navigate customer service in many instances with the implementation of AI assistants that exploit dark patterns made to guide the user to the will of the company and many systems have been designed where the human agent has no authority over the decisions made by the AI automations. I find this despairing and see it as a misuse of the technology, but what can you do? Society will essentially have to learn the hard way since no one is taking into consideration the negative effects that these systems are producing. Eventually there will be a universal understanding of the proper way to use these tools. Until then good luck convincing PayPal or eBay's chat bots to connect you to a live operator for issues not included in their training.