THE AI CONUNDRUM

Book supplement · 2023-01-15

Book Content: You've Got to See It to Believe It

An archived post from the original book website, presented in its historical context.

Here are a few source materials reference in our book that will bring the examples of AI to life.  

First is the AI Kill Bot for CS:GO. Developed by Caleb, in 2020, it needed to fly under the radar and appear human in terms of response time, yet also perform well. The video is from the same computer running the AI, so the number of frames is a bit lower due to a heavily taxed computer processor. 

https://youtube.com/shorts/LrCXUsFjkNw (2020, under 1min)


MONICA debuted at the ANA Masters of Marketing Conference in 2015 in robot form. Here is the Ad Age Story: “HEY, ANA: MONICA'S COMING FOR YOU AND SHE KNOWS WHO YOU ARE, WHAT YOU SPEND, Robot Roaming At Conference Will Promote Free 'Marketing Brain'” Ad Age, Oct 14, 2015, https://adage.com/article/cmo-strategy/marketing-evolution-mma/300894?ttl=1445464524

The YouTube video below shows MONICA, the "Face of Marketing Evolution's ROI Brain" presenting the case study of how MONICA helped Warner Brothers optimize media spend for the movie Creed. 

Watch video on YouTube

https://youtu.be/yR8zBN9Cz7w (publised 2017, 5min)


Google Duplex

Google Duplex is Googles scheduling AI designed to help make appointments. This video, from 2018, shows Duplex making a call to a real salon: 

Watch video on YouTube

https://www.youtube.com/watch?v=bh84hWukDdQ (length: under 2min).

The call is an AI voice and it does an exceptional job impersonating a human. 

We think it is a bad idea to impersonate a human. We strongly advise transparency, and this is not a transparent use of AI for the hair salon receiving the call. Google, as we understand it, is reconsidering how to ensure transparency going forward. 


Google Mistranslation

The mistranslation that opens chapter 8 is from the post "15 Google Translate Fails That Will Make You Never Trust Computers Again" by Steph Koyfman, September 2017. https://www.babbel.com/en/magazine/15-best-google-translate-fails. Our example ranks #2 on the list. The comments to the right explain what the translation should have said.  


It is worth point out the "Rate this translation" which is a good way to get feedback on mistranslations and to provide a level of transparency that the translation is provided by AI. 


We discuss the self-driving cars / autonomous vehicles economic model, which could be massively profitable if the problem is solved. However, given the very nature of Today's AI, we are extremely skeptical that the problems can be solved anytime soon. We contrast Elon Musk's announcement in November 2022, in which he called it "a major milestone" and "available to everyone" followed by Tesla's FSD gone wrong, just hours later:


Watch video on YouTube

https://abc7news.com/video/embed/?pid=11065007

Ultimately, we see Autonomous Vehicles as a "very risky" bet because Today's AI lacks precision and has to operate in an entirely open environment. This is not the ideal dynamics for this generation of AI. 


AI can be few shot learners, meaning with just a few examples, AI can learn how to accomplish the same task. One example is teaching AI to tell dad jokes. Here is Gwern’s site, where he explains the process of teach AI to be as corny as a dad: https://www.gwern.net/GPT-3#dad-jokes.