xAI Grok is on the loose and Heinlein is turning in his grave … but what does it all mean? Is Musk trying to regain the AI advantage after exiting OpenAI not so voluntarily years ago? What kind of a data trove is he sitting on with X and Twitter … or is it just a legal minefield? How does this shift the AI universe if at all?
Few figures have played, and will likely continue to play, as seminal a role in the advancement of AI as Elon Musk.
Each of his companies uses AI in some fundamental way. His history with artificial intelligence policy involves a series of notable events and comments that reflect his evolving relationship with the technology. He cofounded (and exited) the market leader by adoption and revenue. He is pushing the limits of automotive AI and integration with human intelligence at Tesla and Neuralink. And he just announced the new X/Twitter LLM and is heavily promoting it, lauding its real-time capabilities on X last night.
Yet Musk is also one of the people pushing a slowdown in AI the hardest, and has been since 2017. He’s urged the banning of “killer robots”, been clear he believes AI poses a fundamental existential risk to humanity, comparing it to “summoning the demon” at the 2014 MIT AeroAstro Centennial Symposium. He recently made time to talk about AI risk with Rishi Sunak at the UK AI Safety Summit. He was a signatory to the letter. He’s been very vocal about concerns on X. In a 2020 interview, Musk reiterated his worries about superintelligent AI, stating that we are “headed toward a situation where AI is vastly smarter than humans.”
A Complicated History
His history is complicated. He appears to have an underlying ethos of intense caution, but then pushes the technology forward single-handedly in the next breath with a major announcement or development. Some of the major moments:
- Co-founds OpenAI (December 2015): Musk co-founded OpenAI, a non-profit artificial intelligence research company, with the goal of promoting and developing friendly AI to benefit humanity as a whole.
- Warns about AI at the National Governors Association (July 2017): Musk called AI a fundamental risk to the existence of human civilization, urging proactive regulation.
- Leaves OpenAI (February 2018): Musk departed from the board of OpenAI, citing a conflict of interest with Tesla’s own AI development and to avoid any future conflicts as OpenAI moved closer to creating advanced AI technologies.
- Neuralink Public Presentation (July 2019): Musk presented the progress of Neuralink, which aims to build a scalable, high-bandwidth brain-machine interface.
- Push for Tesla’s AI (April 2019 – ongoing): Musk has continued to champion the development of AI in Tesla’s Autopilot and Full Self-Driving (FSD) technologies.
- Comments on AI at the World Artificial Intelligence Conference (August 2019): Musk discussed AI with Alibaba’s Jack Ma, reiterating his concerns but also his optimism about the technology.
- Tesla AI Day (August 2020 and August 2021): Musk showcased Tesla’s advancements in AI and announced the creation of the Tesla Bot, designed to eliminate dangerous, repetitive, and boring tasks.
- Acquires Twitter (October 2022): Musk’s acquisition of Twitter put him at the helm of one of the most influential social media platforms, with a vast amount of data that could be used for training AI systems.
- Announces change to Twitter (now X) TOS (August 2023) to add use of Twitter data, both historical and real time to use in LLM training.
- Launches xAI Grok (November 2023), a GPT LLM trained on the Twitter dataset (exclusively?)
At face value, this history highlights Musk’s active engagement in AI development, his advocacy for its safe and ethical use, and his investments and comments that continue to shape the public discourse on AI. But his actions do not reflect his well articulated “caution”.
More, they are completely inconsistent with Musk’s approach and persona. He leaps at risk and steamrolls through objections and obstacles.
And now, after voicing deep concern, he launches one of the biggest and most lucrative AI datasets in the world: Twitter historical and real-time.
An Unparalleled Dataset
Some context: a year ago, Musk somewhat intentionally acquired Twitter. Twitter’s TOS has always been fairly generous on the side of the company’s ownership of IP. A few weeks ago, Twitter informed its users that the TOS was changing in 48 hours to include , including its dataThere’s never been a dataset like it: multilingual, worldwide, with an enormous range of topics and perspectives, some from experts and officials, conveniently clustered around major worldwide events. Imagine the analysis that is going to be possible. It boggles the mind. It’s also exactly what generative AI needs right now: fresh, massive, new. Musk can supplement with some of the other datasets out there, but he’s the only one who has X, and has it every day. Here’s how it might compare in terms of potential power and utility:
Real-Time Language Usage: Twitter captures language in its most immediate and evolving form. This includes slang, new terminology, and the emergence of real-world events reflected in discourse rapidly.
Diversity of Content: The platform hosts a wide range of topics, opinions, and discussions, representing a diverse cross-section of human thought and communication styles.
Multilingual Data: Twitter’s global user base contributes content in numerous languages, which can be invaluable for training multilingual models.
Public Sentiment and Trends: Tweets often reflect public sentiment, which can be useful for models that need to understand or predict trends and public opinion.
Short-Form Text: Tweets are limited in length, which means the dataset is rich in concise, impactful language use, differing from datasets comprising longer forms of text.
The Twitter dataset is invaluable even before use for LLM training. It narrates the political, cultural and loci of influence of this individuals, multiple societies and communities over an extended period of time. And one person, on whom the mantle of power does not often rest easy, controls it all.
That doesn’t mean there aren’t issues.
One, the data on Twitter is not curated and contains a significant amount of noise, such as spam, misinformation, and non-standard language use, which can pose challenges for training models.
Two, the privacy and ethical issues are unfathomably big. The Twitter TOS are likely to be problematic for Musk out of the gate. Just imagine the potential size of that class-action lawsuit. It’s also unclear how many will continue to use an LLM that explicitly uses “lightly permissioned” non-compensated content for highly lucrative business outcomes.
Three, his positions and his behaviour appear deeply in conflict.
There could be several reasons for the apparent discrepancies. I asked ChatGPT what these might be and this is what it said:
- “Vision vs. Practical Business Decisions: Musk’s caution about AI could be a reflection of his long-term concerns, while his business actions might be driven by immediate market opportunities and competitive pressures. (Possible, but unlikely.)
- Control and Safety: Musk may believe that the risks of AI can be mitigated by building and influencing the technology directly, hence his involvement in AI through companies like Tesla and Neuralink, and now potentially Twitter. (Possible, but again, atypical for him)
- Market Influence: Losing control of the market leader in AI, like OpenAI, might have spurred Musk to pursue his own path in AI through other ventures to maintain influence in the field. (Getting warm!)
- Complexity of AI Governance: The challenges of AI governance are complex, and Musk’s actions might reflect a strategy to be a part of the conversation and solution, which sometimes requires being actively involved in AI development. (Very doubtful)
It’s important to note that business leaders do often have to balance idealistic caution with practical business interests, leading to actions that can seem at odds with their stated beliefs. Without a clear insight into Musk’s intentions and plans, it’s challenging to definitively interpret his actions.”
A Striking Inconsistency
But we must. How do we make sense of Musk’s positions? One, we can say he is conflicted. Two, we can say he is opportunistic. Three, we can say he is ruthless and risk-interested enough to publicly caution against a technology so he can gain an advantage.
He’s asked the industry to slow down. But he is accelerating.
AI work at Tesla and Neuralink, some of the most advanced in the world, is not slowing down. It is massively accelerating at X. What can we start to take away from all of this?
All the policies and summits are utterly meaningless right now if one guy decides he wants to push. And he is pushing. He can supplement the holes in the Twitter dataset, and he doesn’t care about laws. Meanwhile, competitors see nothing but regulatory headaches Musk himself will never submit his companies to.
The rocketship traction of ChatGPT has not helped. Musk likes to win. He is not happy about a company he lost control of gaining market dominance. This is true going back to back to PayPal. Is the announcement that X will now be a financial platform a coincidence as well? There’s something to be said for exercising petty grievances as a path to profitability. If anyone can make that work as a business model, it’s Elon.
The possibility Musk is intentionally distracting the world while pushing ahead with full force himself is high. Human ingenuity is incredible, but greed and lust for power have always found a way to co-opt tech to their own advantage. We know he can be petty. Petty with this kind of power is frightening.
All the risks with Musk increase daily. His coziness with right wing thinkers and dictatorial thinkers, plus the occasional fascist; his lack of concern for potential harm to specific vulnerable groups belies any apparent other concern for humanity in general. This plus AI, his impetuousness, his ego, his influenceability, his mental health and sleep, may make Musk himself the single biggest risk AI holds over us.
A risk it is unclear how to manage.
Written with support from ChatGPT
