Google Duplex's Failure Should Teach us Something about A.I. Adoption in the real world

A.I Alignment actually matters

Hey Guys,

A.I. products are really hard to get right. ChatGPT has brought up a host of A.I. alignment (A.I. ethics) questions to mind. This post by Ralph Aboujaoude Diaz gives some important context and logic.

Most of the hype around ChatGPT isn’t rational and certainly isn’t practical, and we’ve seen that before. Anyone remember Google Duplex? Google Duplex had the same major ethical A.I. alignment issues. Its result? Abject failure, a common thing with Google it turns out.

So even as Amazon Alexa is at the brink of failure and Google Duplex has failed, we hype the next thing.

TechCrunch broke the story on December 2nd, 2022 that Google is shutting down Duplex on the Web, its AI-powered set of services that navigated sites to simplify the process of ordering food, purchasing movie tickets and more. According to a note on a Google support page, Duplex on the Web and any automation features enabled by it will no longer be supported as of this month.

The Year was 2019

I too remember the excitement of watching the Google Duplex Demo already four years ago.

Chatbots, are hard guys.

The so-called ‘Duplex’ feature of the Google Assistant was shown calling a hair salon to book a woman’s hair cut, and ringing a restaurant to try to book a table — only to be told it did not accept bookings for less than five people. Supposed to be useful? In reality, not so much.

Google introduced Duplex on the Web, an outgrowth of its call-automating Duplex technology, during its 2019 Google I/O developer conference. It felt wonderful, a bit like our December honeymoon with ChatGPT in 2022.

The hype was glorious, the glow of another supposed “win” for A.I. After robo-taxis and the disgrace of Alexa for Amazon, I think we know the score on A.I. and hype. Hype is actually a bad sign for real-world adoption. Generative A.I. eh? Not to mention in the case of Duplex, Google is just really bad at product-marketing, I have no idea why. We saw the same thing with Stadia and dozens of other products.

Microsoft Cortana in mostly unused on Windows, let’s just admit it. CAn they find commercial utility with ChatGPT like technology trained on RLHF on GPT-4, sure they can, but it won’t be easy. This ain’t no Teams vs. Slack, where we have thousands of partners and sellers to unfairly compete in a monopoly Capitalism.

That Google shut down Duplex is a tragedy of a lack of real-world adoption, yet nobody really noticed. We were all busy talking about ChatGPT, what it could and couldn’t do and the gimmicks it would lead to.

Viral Tweets on ChatGPT are Like Misinformation

Most of the viral tweets about ChatGPT, here and here, are founded on ill-founded conceits with little grounding in reality. Remember folks, misinformation really does spread more virally. Such is the case with A.I. adoption as well.

ChatGPT is a bit like a publicity stunt for OpenAI, that now has a history of not being very open.

The new decentralized A.I. research collectives really actually care about A.I alignment you see.

Luca Soldaini ✈️ EMNLP 2022 @soldni

Frankly #ChatGPT makes me so sad and disillusioned with NLP research. It's not the model not being available; it's not the fact that is trained on unknown data and methods; it's not the clear commercial priorities and marketing. But the combo of all three is such a blow 😔

Google Duplex also felt a bit wrong, not just on ethical grounds but fundamentally a bit unhinged. In many respects I’m getting the same feeling about ChatGPT and the Generative A.I. hype. It’s almost as if crypto trained us on Twitter to be dishonest about how we react to things.

FOMO was mass conditioned into us like a mob.

I would myself prefer to cover robotics or some startup, but it needs to be said. Chatbot demos aren’t particularly good at predicting the future of A.I. I also remember covering Microsoft Tay back in 2016.

While Microsoft has significant advantages with OpenAI, OpenAI’s foundational models aren’t likely to be the eventual winners, they aren’t grounded in A.I. alignment enough. OpenAI doesn’t represent the majority of serious A.I. researchers in terms of ethical leadership and how it thinks of A.I. trust, safety and transparency. It is instead grounded in commercialization.

I think it’s safe to conclude in 2022 that Meta, Google or Microsoft aren’t likely good actors in bringing us A.I. Many of the research projects of Meta, are deeply flawed in a lack of due diligence in A.I. alignement.

In fact the dystopian side of Generative A.I. may even be more significant than the benefits. The A.I. risk may outweigh the benefits, but hardly anyone is talking about that. Why do you suppose that is?