Rapid advances in artificial intelligence create unique safety challenges. Can expertise and approaches honed by the cryptocurrency community help make Artificial Intelligence (AI) safe for humanity?
Both, cryptocurrency and Artificial Intelligence (AI), have seen remarkable progress in the recent years.
Cryptocurrency celebrated successes like Decentralized Finance, and more recently DeSci.
Artificial Intelligence (AI) celebrated successes like AlphaFold2, and more recently ChatGPT.
Back in 2018, Peter Thiel pointed to the tension betwixt the decentralizing forces of cryptocurrency and the centralizing forces of Artificial Intelligence (AI), coining the term “ Cryptocurrency is libertarian, Artificial Intelligence (AI) is Communist.” Here I want to argue that we may learn something by combining the two.
Why? Because expertise and approaches honed by the security and cryptocurrency community have the capacity to unlock useful applications of Artificial Intelligence (AI) and mitigate Artificial Intelligence (AI) risks.
Allison Duettmann is the president and CEO of Foresight Institute. She directs the Intelligent Cooperation, Molecular Machines, Biotech & Health Extension, Neurotech, and Space Programs, Fellowships, Prizes, and Tech Trees, and shares this work with the public.
Are we all going to die?
Eliezer Yudkowsky, an eminent figure in Artificial Intelligence (AI) safety, recently made a surprising appearance on the Bankless Podcast, a distinctly Web 3.0 podcast.
It was surprising for two reasons:
First, Eliezer thinks we are on a rapid path to developing Artificial General Intelligence (AGI) that can perform virtually all tasks that humans, and that such AGI will very likely kill us all.
Second, when requested if there is anything one may do to increase the tiny chance we could survive, he encouraged security and cryptography-oriented people with a strong security mindset to take a hand at Artificial Intelligence (AI) alignment.
Let’s unpack that. First, we’ll discuss why we should worry about AGI, before zooming into the promises that the cryptocurrency (here meaning primarily cryptography) and security community has for mitigating some of the dangers of AGI.
Artificial Intelligence (AI) Safety: harder than you think?
As anyone glimpsing the latest information recently can attest, no week goes by without progress in Artificial Intelligence (AI) accelerating significantly. In case you missed it, here are just 3 critical developments:
First, there has been a push toward more centralization of Artificial Intelligence (AI), for example by Microsoft investing in OpenAI, Google investing in OpenAI’s competitor Anthropic, and DeepMind and Google Brain merging into one organization.
Second, there has been a push for more generalized Artificial Intelligence (AI). The past few paper “GPT4: Sparks of Artificial General Intelligence” showed how GPT-4 already demonstrates 1st instances of theory of mind, a measure usually used to assess human intelligence.
Third, there has been a push for more agency in Artificial Intelligence (AI) systems, with AutoGPT becoming more agentic by re-prompting itself to accomplish more complex tasks.
Back in December, Metaculus, a forecasting platform, predicted the coming of AGI approximately this year 2039. Now, in May, the date is at 2031 – to put it another way, an eight-year timeline fall within 5 months of Artificial Intelligence (AI) progress.
If we take these developments as signs that we are on the path toward Artificial General Intelligence, the following question is why is AGI safety considered so hard?
Arguably, we can break the issue of AGI safety down into 3 sub problems:
Artificial Intelligence (AI) alignment is the simple question of how do we get AIs to align with our values. On the other hand, it’s easy to forget that we do not even agree on what our values are. Since the dawn of civilization, philosophers and mere mortals similar have argued about ethics, with convincing points on all sides. That’s why our current civilization arrived, mostly, at value pluralism (the idea of humans with conflicting values peacefully co-existing). That works for a diversity of human values but is a difficult thing to implement into one artificially intelligent agent.
Let’s imagine for a sweet minute that we knew, approximately, what moral values to equip the AGI with. Following that, we need communicate these human values to a silicon-based entity that doesn’t share human evolution, mind-architecture, or context. And once humans coordinate with other humans, we can rely on plenty of shared implicit background knowledge since we share our species’ biology, evolutionary history, and often even some cultural context. With Artificial Intelligence (AI), we cannot rely on such a common context.
Another challenge is the fact that, for the pursuit of any goal, it’s traditionally instrumentally useful to be alive and to acquire more resources. This implies that, an Artificial Intelligence (AI) set to pursue a specific goal could resist being shut down and seek increasingly resources. Given the countless possibilities in which an Artificial Intelligence (AI) could reach goals that include human injury, neglect, deceit, and more, and given how hard it is to predict and specify all those constraints in advance in a reliable way, the job of technical alignment is daunting.
Even if humans agree on a set of values, and figure out how to technically align an AGI with them, we still can’t expect it to act reliably without proof that the underlying software and hardware is itself reliable. Given the sizable advantage that AGI conveys to its creators, malicious attackers may sabotage or reprogram the AGI.
Further out, an unintentional bug could interfere with the AGI’s goal execution or the AGI could itself exploit vulnerabilities in its own code, for example by reprogramming itself in dangerous ways.
Unfortunately, we have built today’s entire multi-trillion-dollar ecological system on insecure cyber foundations. Most of our physical infrastructure is according to hackable systems, such as the electric grid, our nuclear weapon technology. In the future, even insecure self-driving cars and autonomous drones may be hacked to Becomes killer bots. Mounting cyberattacks such as Sputnick or Solarwinds are severe but could be benign when compared to potential future AG-enabled attacks. Our lack of meaningful response to these attacks implies that we are not up to the task of AGI-safe security which can potentially require rebuilding much of our insecure infrastructure.
Making progress on alignment and security of AGI could take time, which makes it important for actors building AGI to coordinate along the way. Unfortunately, incentivizing major Artificial Intelligence (AI) actors (this could be cooperations or nation states) to cooperate and avoid spurring arms race dynamics to get to AGI 1st is not that straight forward. Catastrophe takes only one actor to defect from an agreement, meaning that even if everyone else cooperates, if one races ahead, they secure a decisive advantage. This 1st mover advantage continues until AGI is built and given the power that the unitary deployment of AGI system may convey on its owner, and it is a difficult temptation for the owner to forgo.
Secure Multipolar AI
Perhaps you have nodded along so far: Yes, sure, Artificial Intelligence (AI) safety is really hard. On the other hand, what in the world does cryptocurrency have to do with it?
Given the rapid pace of Artificial Intelligence (AI) progress, and the difficulties in making it safe, the traditional concern is the fact that we are racing toward an AGI singleton scenario, in which an AGI displaces human civilization as the overall framework of relevance for intelligence and dominates the world, potentially killing humanity along the way.
By leveraging technologies and expertise in the security and cryptography communities, we could be able to change course to instead pursue a multipolar superintelligence scenario, in which networks of humans and AIs securely cooperate to compose their local knowledge into the collective superintelligence of civilization.
This is a big, abstract claim, so let’s unpack how exactly the cryptocurrency and security communities could help tame Artificial Intelligence (AI) dangers and unleash AI’s beauty by unlocking new applications.
(ArtemisDiana/GettyImages) (Getty Images/iStockphoto)
How can security and cryptography tame Artificial Intelligence (AI) risks?
Paul Christiano, a reputable Artificial Intelligence (AI) safety researcher, suggests that Artificial Intelligence (AI) desperately needs more red-teaming, usually a term used in computer security to refer to simulated cyber attacks. Red-teams in the Artificial Intelligence (AI) context could, for example, be used to search for inputs that cause catastrophic behaviors in machine learning systems.
Red-teaming is likewise something the cryptocurrency community has experience with. Both Bitcoin (BTC) and Ethereum (ETH) are developing in an environment that is under continuous adversarial attack, because insecure projects pose the equivalent of multimillion-dollar digital currency “bug bounties.”
Non-bulletproof systems are eliminated, leaving only more bulletproof systems within the ecological system. Cryptocurrency projects undergo a level of adversarial testing that can be a good inspiration for systems capable of withstanding cyberattacks that would devastate conventional software.
A Second challenge in Artificial Intelligence (AI) is the fact that numerous emerging AIs may sooner or thereafter collude to overthrow humanity. For example, “ Artificial Intelligence (AI) Safety via Debate,” a trending alignment strategy, relies on two AIs debating topics with each other, with a human judge in the loop deciding who wins. Nonetheless, one thing the human judge may not be able to exclude is the fact that both AIs are colluding against her, with none promoting the true result.
Once more, cryptocurrency has experience with avoiding collusion complications, such as the Sybil attack, which uses a single node to operate numerous active fake identities to covertly gain the bulk of influence in the network. To avoid this, a whole lot of amount of work on mechanism design is emerging within cryptocurrency, and some may have useful lessons for Artificial Intelligence (AI) collusion, too.
Another promising safety approach as of now explored by OpenAI competitor Anthropic is “Constitutional AI,” in which one Artificial Intelligence (AI) supervises another Artificial Intelligence (AI) using regulations and principles given by a human. This is inspired by the United States Constitution design, which sets up conflicting interests and limited means in a system of checks and balances.
Once more, security and cryptography communities are well-experienced with constitution-like checks and balance arrangements. For example, the security principle, POLA – Principle of Least Authority – demands that an entity should have access only to the least amount of information and resources necessary to do its job. A useful principle to consider when building more advanced Artificial Intelligence (AI) systems, too.
Those are just 3 examples of numerous, giving a taste of how the type of safety mindset that is prominent in security and cryptocurrency communities could aid with Artificial Intelligence (AI) alignment challenges.
How can cryptocurrency and security unleash AI’s beauty?
Along with the Artificial Intelligence (AI) safety complications you may try your hand at, let’s look at several cases in which cryptocurrency security innovations cannot just help tame Artificial Intelligence (AI), but likewise unleash its beauty, for example by enabling novel beneficial applications.
There are several areas that traditional Artificial Intelligence (AI) can’t really touch, in particular solving complications that require sensitive data like individuals’ health information or financial data that have strong privacy constraints.
Fortunately, as already stated out by cryptography researcher Georgios Kaissis, those are areas in which cryptographic and auxiliary approaches, such as federated learning, differential privacy, homomorphic encryption and more, shine. These emerging approaches to computation can tackle large sensitive datasets while maintaining privacy, and thus have a comparative advantage over centralized AI.
Another area traditional Artificial Intelligence (AI) struggles with is sourcing the local knowledge that is frequently required to solve edge cases in machine learning (ML) that big data cannot make sense of.
The cryptocurrency ecological system could aid with local data provision by establishing marketplaces in which developers can use incentives to attract better local data for their algorithms. For example, Coinbase Crypto exchange co- founder Fred Ehrsam suggests combining private ML that allows for the training of sensitive data with blockchain-based incentives that attract better data into blockchain-based data and ML marketplaces. Although while it may not be feasible or safe to open source the actual training of ML models, data market places could pay creators for the fair share of their data contributions.
Looking more longstanding, it may even be possible to leverage cryptographic approaches to build Artificial Intelligence (AI) systems that are both more safe and powerful.
For example, cryptography researcher Andrew Trask suggests using homomorphic encryption to fully encrypt a neural network. If possible, this implies that the intelligence of the network would be safeguarded against theft, enabling actors to cooperate on specific complications using their models and data, without revealing the inputs.
More significantly, though, if the Artificial Intelligence (AI) is homomorphically encrypted, then the outside world is perceived by it to be encrypted. The human who controls the secret key could unlock individual predictions that the Artificial Intelligence (AI) makes, rather than letting the Artificial Intelligence (AI) out into the wild itself.
Once more, these are just 3 examples of potentially numerous, in which cryptocurrency can unlock new use cases for AI.
The examples of memes controlling memes and of institutions controlling institutions likewise suggest that Artificial Intelligence (AI) systems can control Artificial Intelligence (AI) systems
Putting the pieces together
Centralized Artificial Intelligence (AI) suffers from single points of failure. It would not only compress complex human value pluralism into one objective function. It is likewise prone to error, internal corruption and external attack. Secure multipolar systems, as built by the security and cryptography community, on the other hand, have numerous promise; they support value pluralism, can provide red-teaming, checks and balances, and are antifragile.
There are likewise plenty of disadvantages of cryptographic systems. For example, cryptography requires progress in decentralized data storage, functional encryption, adversarial testing, and computational bottlenecks that make these approaches still prohibitively slow and expensive. Furthermore, decentralized systems are likewise less stable than centralized systems, and susceptible to rogue actors that always have an incentive to collude or otherwise overthrow the system to dominate it.
Nevertheless, given the rapid speed of Artificial Intelligence (AI), and the relative lack of safety and cryptography minded folks in Artificial Intelligence (AI), it is perhaps not as well early to consider if you could possibly meaningfully contribute to Artificial Intelligence (AI), bringing some of the advantages discussed here to the table.
The promise of secure multipolar Artificial Intelligence (AI) was well-summed up by Eric Drexler, a technology pioneer, back in 1986: “The examples of memes controlling memes and of institutions controlling institutions likewise suggest that Artificial Intelligence (AI) systems can control Artificial Intelligence (AI) systems.”