Social Search on the Blockchain — Why we need it
Finding dates, doctors, and difficult answers through collective intelligence
“Who’s a good dentist in San Francisco?”
“Who’s the right guy/girl for me?”
“Where’s the bug in my code?”
“What job should I do next?”
Google is great and will keep getting better with time, but try asking these questions and you might find the answers less than satisfactory. Sure, there are separate databases of information for each information vertical — LinkedIn and Angel List have jobs data, Yelp can offer restaurant recommendations, and a host of dating apps and healthcare startups will promise you personalized matches for your romantic life and next doctor visit — but these search engines and apps can’t solve the fundamental problems that make these questions difficult to answer in the first place:
- Not all information is online. Much is locked away in human memory, sitting on a Word Doc in “My Documents”, or scribbled somewhere in a notebook.
- Some decisions are hard for present-day algorithms and require human experience. This may be because of ambiguous data, the lack of historical trends, the need for high levels of personalization, or because the objects being evaluated and compared in the search process are intrinsically experience goods rather than search goods, defying if-then heuristics and keyword lookup. “People are experience goods,” a 2008 research paper by Frost, Chance, Norton & Ariely elaborates on this dichotomy in the case of online dating.
AI and machine learning can attempt to solve Problem 2 by allowing for fuzzy logic learned from historical data, but will hit a bottleneck with Problem 1.
Our most powerful ML approaches today are ‘supervised’, meaning that they need to be trained on large, labeled data sets before they can start making reliable inferences. In the narrow verticals where data has become increasingly available, the ML research community and a select group of large tech firms have been able to build AI that achieves better-than-human performance (e.g. Microsoft with image classification), but for other areas, e.g. debugging code and teaching someone how to do something, it will be a while before our computational approaches for inference can compete with the enigmatic mix of Bayesian program, neural network, and long-term memory that is the human brain.
Social search requires Trust and Incentives — difficult to create and scale
The idea of crowdsourcing search queries and using human expertise to filter and rank relevant results has existed for years, and tens of startups, particularly in the 2008–2016 timeframe were conceived and shut down in the pursuit of a human search engine for consumers at scale.
Aardvark started the trend in 2007 — a team of ex-Googlers built a real-time q&a system that allowed information seekers to broadcast questions to friends and friends-of-friends via their IM platforms, email, or the Aardvark website. The product went viral, was acquired by Google in 2010, and then shut down. Then came PeerPong (shut down in 2011), Mahalo (shut down in 2014), ChaCha (shut down in 2016), and many others, all vying to be the platform that cracks social search but struggling with either low adoption from consumers, poor demand from responders, or unprofitable unit economics.
The same things that make social search awesome — the humans and the network effects — are also what make it so challenging:
- Humans require incentives — to join the platform as first adopters when the value to them of the network is near zero, and to continue to respond to search queries as the network grows.
- Humans make mistakes — and can provide false answers to search queries innocently or maliciously. Given this human flaw, a human search engine needs to create trust in its ability to weed out bad answers. A simple solution seems to be peer review, but reviewers are human and will need incentives to review as mentioned above, though some might disagree (e.g. see comments from Silvio Micali, the MIT professor behind Algorand, on the potential down side of incentives).
Social information-exchange platforms like Quora (> 100M users), StackOverflow (> 7M users), and Brainly (70M users) have amassed large user bases by focusing on free, quick, and simple Q&A, and by using Reputation (see Stackoverflow’s “Theory of Moderation”) as the source and enforcer of Trust, but it is important to keep in mind that the type of q&a they support is only a tiny piece of Social Search overall. Questions such as “Whom should I date” would yield equally unsatisfactory and excessively generalized results on these platforms, and if users start asking time-consuming queries (e.g. “Can you rank this list of companies for me based on what is the best employer for me”), responders will start asking for incentives beyond mere social validation and reputation.
Blockchain solves the problem of Trust and Incentives
We believe that in the specific but vast realm of Social Search, the blockchain provides important benefits that would allow a global-scale social search engine to succeed on the blockchain where it would have otherwise struggled or failed on traditional architectures.
- Ease of creating financial incentives via automatic, trustless, p2p payments: Anybody can transact with anybody, regardless of geographical location or currency differences, without an intermediary. Payments are direct and micro-payments (small denominations) are easy. Building apps with frictionless p2p financial exchange off the blockchain, though not impossible, has proved difficult. Presenting paywalls to users via 3rd party payment gateways for every transaction is hard from both a development and UX perspective, and even mobile wallets (PayPal, PayTM, WePay) don’t provide the global access and customizability that blockchain transactions with smart contracts provide.
- No single point of control or censorship for search results: The centralized search apps of today modify search results all the time — e.g. re-ordering results on Google.com to favor those that are ‘sponsored’ and backed by Ad dollars, or altering / suppressing / inserting results on the Facebook news feed. The blockchain makes such result modifications nearly impossible and permits censorship only if it is built into the dApp logic to begin with.
- Ease of creating trust by incentivizing removal of bad answers: Programmable smart contracts on the Ethereum blockchain allow for customizable transaction logic difficult to create with traditional payment mechanisms, for example:
A answers question asked by Q.
V verifies question answered by A.If A’s answer is right: automatically transfer $$ from Q to A
If A’s answer is wrong: do not pay A, Q keeps its money
Either way: Q pays V a verification fee
A “protocol” for social search on the blockchain
If we believe in the benefits of the blockchain mentioned above, the next question to be answered becomes — how do we go about building something on the blockchain that enables the various use cases of social search. Each of the four questions at the start of this article will require a different app on the blockchain (crowdsourced matchmaking vs medical advice vs crowdsourced job-search), but they will require a common mechanism by which incentives and trust are created and managed across various participants.
Much in the same way that network protocols like TCP/IP and HTTP enable multiple use cases and web applications to be built on a foundation of client-server communication, protocols for the blockchain allow for various dApps (decentralized apps) to be created with shared logic taking care of the messy stuff.
Will Warren from the 0x project offers an explanation for blockchain protocols in his writeup on app tokens vs protocol tokens that we particularly like (“cryptoeconomic protocols create financial incentives that drive a network of rational economic agents to coordinate their behavior towards the completion of a process”), and Vinay Gupta offers a brilliant historical perspective on the evolution of and motivation for protocols, citing benefits such as enabling interoperability between multiple apps (e.g. SMTP, IMAP, POP power email across various email apps) and shared network effects across multiple use cases.
Further Exploration: 5 Ethereum apps for social search in 5 days
At Proffer, we have been thinking about and working on what a protocol for Social Search would look like, and we leveraged the Token Hackathon run by Coinbase as an opportunity to build out several dApps bound by common logic for trust and incentives. We built 5 apps in 5 days and were pleased to hear that our apps collectively won the hackathon.
In the coming days, we’ll publish 1 post for each of those apps, describing how a protocol for Social Search on the blockchain would enable each use case. My teammate Sinchan Banerjee will also publish a post that digs deeper into the entities and information flow that a search on Proffer will involve.
This is just a start — stay tuned for updates and more importantly, for social search coming to a blockchain near you.
