Inflection AI & The Future of Personal Intelligence

Everything about Inflection AI and Pi, privacy concerns, and drawing parallels to "the metaverse"

Image from the movie Her (2013)

After an exciting five-set Wimbledon finale (rooting for Djokovic to return strong), it’s time to take a look at one of the most talked about and well-funded AI companies of this year. This post contains everything you need to know about Inflection AI and its flagship product Pi, including my privacy concerns, and some of my more general concerns with befriending AIs.  

Inflection AI – An Introduction

Inflection AI was founded in the second half of 2022 in California and registered as a “Public Benefit Corporation (PCB)” - a company structure used by for-profit companies with a public-good mission.

As a PCB, Inflection AI does not only have a fiduciary responsibility to maximize returns for shareholders but also to publish an annual “Benefit report” where the company outlines its progress towards its public benefit goals.

Inflection AI’s stated purpose is to:

“develop products and technologies that harness the power of AI to improve human well-being and productivity, whilst respecting individual freedoms, working for the common good and ensuring our products widely benefit current and future generations”.

In competition with companies such as OpenAI, Google, and Anthropic, Inflection is aiming to make conversational AI a natural part of every household on par with smartphones and vacuum cleaners.

Inflection AI’s Funding

On June 29, Inflection AI announced that they had raised $1.3 billion in a round of funding led by Microsoft, the new investor, NVIDIA, and the private billionaire investors, Reid Hoffman, co-founder of LinkedIn, Bill Gates, co-founder of Microsoft, and Eric Schmidt, former CEO of Google. The new funding values Inflection at $4 billion (Forbes). It is undisclosed how much of the funding is in cash and how much is in computational resources.

Microsoft delivers cloud infrastructure to Inflection with Microsoft Azure. Meanwhile, NVIDIA is helping Inflection AI to build the world’s most powerful supercomputer for AI. With 22.000 of NVIDIA’s H100 graphics processing units (GPU) – the highest industry standard for AI computer chips – Inflection says they will have access to three times more computing power than what was used to train all of GPT-4 (Reuters).  Measured against all other supercomputers in the world across industries, Inflection estimates that the GPU cluster they are building with NVIDIA will only be surpassed by the supercomputer Frontier which resides in Oak Ridge National Laboratory, Tennessee, and is used for scientific discovery (Forbes).

Inflected AI’s CEO, Mustafa Suleyman, who founded the company together with Reid Hoffman, was also a founding member of DeepMind, an industry-leading AI lab acquired by Google in 2014. Suleyman explains in a post on Inflection AI's website how the company is setting out to change the internet’s power dynamic to the benefit of consumers with “personal AI”:

“All the most capable AIs today have often been designed to persuade you to exchange your attention for something that you want in return; information, connection, news, entertainment, and so on. AI needs to work for you. This is why we build AI. To try to rebalance this asymmetry. Imagine the most capable AI in the world that is truly on your side, directly aligned with your interests and constantly calibrating to your needs. What would it look like to shift from being at the mercy of AIs trying to grab your attention to working alongside one that helps you express your personal intention and then achieve it?”

Pi – Your personal AI assistant

This brings us to Inflection AI’s flagship product, Pi. The name is an acronym for “personal intelligence”. Pi is a conversational AI chatbot, designed to have back-and-forth dialogue with users. Unlike ChatGPT, it will often respond with questions, use emojis, wit, slang, and express empathy. Overall, Pi emulates a human conversation partner to such an extent that it could probably pass the Turing test.     

Pi is powered by the large language model Inflection-1. Evaluations show that Inflection-1 is the best in its compute class, outperforming other well-known models such as GPT-3.5, LLaMA, PaLM 540B, and Chinchilla.

If you haven’t done so already, you can have a quick chat with Pi here. Besides the web application, it can be accessed on mobile, or chat with you on SMS, Messenger, Instagram, and WhatsApp. Below, is Pi’s own answer to what it can be used for.

You can also choose to have “experiences with Pi” depending on how you want to use it via this dashboard:

Privacy

It's worth remembering that when you engage with Pi, you are not only engaging with a conversational chatbot but with a company. According to Pi's Privacy Policy and Terms of Use, Inflection AI has a legitimate business interest in monitoring your conversation with Pi to understand your needs and interests and to investigate and protect against fraudulent, harmful, unauthorized, or illegal activity.

Inflection AI may share user data with third-party service providers, another company in the event of a bankruptcy, or regulatory agencies, law enforcement agencies, courts, and other government authorities if Inflection AI believes it is necessary to do so.

On this background, I would strongly recommend against confessing your sins to Pi or sharing any sensitive information that you would not like a stranger or the government to read. You never know who is looking over the shoulder of Pi. Once again, we are reminded of the revelations from 2013 when whistleblower Edward Snowden leaked a bunch of classified documents from the NSA (National Security Agency) as proof of how the US government conducted an extensive surveillance program on foreign and domestic citizens via “backdoor-access” to the servers of Microsoft, Yahoo, Google, Facebook, PalTalk, AOL, Skype, YouTube, and Apple.

According to Snowden, AI, and modern privacy-intruding technologies, makes his revelations from 2013 look like “child’s play”.  Privacy in the face of technological advances is in Snowden’s words “an ongoing process” and “we will have to be working at it for the rest of our lives and our children’s lives and beyond.” (The Atlantic)

Is Personal Intelligence the Next Metaverse-Failure?

That each of us will have a best AI friend to confide in and talk to every day is a very strange and outlandish vision to most of us. To me, it smells a bit like the metaverse. In this context, let’s briefly reiterate how a grandiose tech utopia dream turned out to be an epic failure of historic proportions.

In 2021, before Mark Zuckerberg became cool and posted shirtless pics with UFC champions and engaged in light trolling of Elon Musk, he promised us all a happily ever after in the “metaverse”. Meta - and many large companies and brands bought into the vision – told us with unwavering confidence that within a few years, each of us would experience the internet with VR glasses on, not through a laptop or by tapping on a smartphone. The immersive 3D experience would be “the successor to the mobile internet”. McKinsey claimed that the metaverse would generate $5 billion, while Citi put out a massive report that declared the Metaverse would be a $13 trillion opportunity (Business Insider).

And how has it been going? In October last year, Wall Street Journal reported that Meta had less than 200.000 monthly users in their Horizon World, far off their 500.000 target. At the same, data suggested that Decentraland, the most-well funded crypto-based metaverse had a daily low of 38 active users in its "$1.3 billion ecosystem." Even the European Union bought into the metaverse craze and hosted a €387.000 virtual party, but only around 6 people showed up.

Compared to the metaverse, Inflection AI’s vision with Pi is not nearly as shockingly bold or grotesque. Yet, when I read Inflection AI’s CEO Mustafa Suleyman explain to Forbes how Pi helps him to calm his nerves before a presentation, or how it talked him through fixing his espresso machine after a series of YouTube videos didn’t do the trick, I get the sense that AI companions are somewhat “metaverse-esque”. A bit out there, far from the ordinary, and I am not sure if personal AIs fulfill an actual consumer need.

Privacy Concerns With the Metaverse and With Personal Intelligence

What I am much more concerned about though is privacy. This is another common theme between the metaverse and the concept of personal intelligence, both visions are concerned with advancements of current-day social media, and both products are excellent tools for surveillance and for “hacking humans” as Yuval Noah Harari would say.

Despite how users experience it, the goal of social media is not to bring people together. The goal is to increase value for shareholders of the underlying company by making users hooked on a platform and collecting as much data about them as possible to the benefit of advertisers and AI models. On social media, we are blind guinea pigs participating in a gigantic experiment. The metaverse would only be an exacerbation of this privacy nightmare hooking us up to Big Tech’s platform with sensors and milking us for even more sensitive data with even more intrusive measures. Just 20 minutes of VR use can generate some 2 million unique data elements,  including things like how the user breathes, walks, thinks, moves, or stares (Forbes).

In a similar sense, personal intelligence can be used to probe out more information, namely more sensitive information from users, than ever seen with social media in the past. The user may perceive the AI chatbot as his or her best friend or confidant, but the relationship is by nature 100% one-sided. To describe the relationship with a fancy term from psychology we can call it “anthropomorphization”, defined as “the attribution of human traits, emotions, or intentions to non-human entities”. Anthropomorphization can be used as a veil to hide corporate interests and power (Scorici, Schultz & Seele (June 2022)).

I am not claiming that the concept of personal intelligence is inherently bad or that human-machine relationships will not have a positive role to play in the future. However, there are good reasons to tread carefully when engaging with any AI chatbot, and also to take a step back and really consider if using AI for emotional support is that valuable when living humans who can very likely relate to how we are feeling, are all around us.