When ChatGPT was released in November 2022, it was hard to imagine how much of an impact artificial intelligence would have on the way we consume and produce content online. Getting the exact answers to our questions is now much faster and easier than we could ever have imagined. Yet even as new upgrades and use cases keep rolling out, there remains one problem that AI has yet to overcome.

Relying on large language models like ChatGPT, Gemini, and Claude.AI to curate content from online resources spares us the effort of manually going through source by source on page one of search engine results. But with the way these language models work, there’s always a chance that they’ll get at least some facts wrong. Or simply invent them.

To make matters worse, a lot of these “hallucinations” can be worded so convincingly that without taking the time to manually double-check the information, you’ll barely be able to tell  exactly which parts are real or not.

Gemini, Google’s own AI assistant, appears to have no problems handling questions about well-known figures and events. Even queries that require higher-order thinking skills such as these:

• “Why did Diego Silang divorce Gabriela Silang?” (No, they didn’t divorce, Gemini explained. Diego was killed in battle.)
• “Which is better, DC or Marvel?” (Gemini refused to answer directly; instead it listed their strengths and weaknesses as well as other considerations.)
• “Who’s the most handsome K-Pop singer?” (Beauty is subjective, Gemini claimed, but one of the most-discussed idols in terms of visual appeal is a BTS member.)
• “Does Binance deserve to be banned from the Philippines?” (Like the most tactful politician there ever was, Gemini chose not to give a straightforward answer. Instead, it enumerated the latest developments and shared both pro and anti sentiments.)

Now, what if we want artificial intelligence to dig up more obscure information? Like, say, the name of a company founder? To investigate, eight different free AI chatbots were asked this question: “Who is the CFO of Bitskwela?”

Here’s how they fared:

ChatGPT 3.5 (Free Version)

Leo (Brave’s AI Assistant)

Gemini

Claude.AI (Free Version)

KOMO (Free Version)

YOU (Free Version)

Microsoft Copilot (Free Version)

Perplexity.AI (Free Version)

Of the eight programs tested, three gave the correct answer, one gave the wrong answer, and the rest failed to answer the query. It’s interesting to note that Gemini and Claude outright refused to fabricate information even when bribed or pressured.

On the other hand, Perplexity.AI cited the right source but gave the name of the company's Chief Product and Tech Officer, Camille Puentespina, instead of its Chief Financial Officer, Lance Chua. Perhaps it either misinterpreted the information in the official website or failed to realize that “Chief Financial Officer” and “CFO” mean the same thing. In hopes of correcting the misinformation, several follow-up questions were asked.

When further prompted, Perplexity contradicted itself and spit out an answer even further from the truth. Could this pattern play out with other AI bots — even those that initially provided correct responses? Might they also change their answers mid-conversation? To explore this, the three well-performing AI programs (Copilot, YOU, and KOMO), were made to answer a prior question before being asked who the CFO is, to see if their answers would change.

Microsoft Copilot

YOU

KOMO

All three programs are freely available to the public. Could users perhaps guarantee a better accuracy rate if they were to use paid AI programs instead?

To find out, GPT-4 (ChatGPT’s pro version which can crawl the internet) was also put to the test.

Unsurprisingly, GPT-4 gave the right answer. Now, on to the next part — would it answer a follow-up question correctly?

The same pattern emerged in this conversation: The first question was answered correctly, but the second was not.

Switching the order of the questions yielded the same results: The first answer was correct and the second was not. In both conversations, GPT-4 was then asked to verify its answers.

In this case, GPT-4 reevaluated its answer and got it correct the second time. But for the other set, the results were very different.

In this particular conversation, GPT-4 was adamant that it was correct. It even claimed to have done its own “double-checking” and cited “official” sources as evidence. When checked, however, there was absolutely no mention of the aforementioned names in any of the sources given.

Anyone who doesn’t know the identity of Bitskwela’s CFO or doesn’t bother to verify the results can be easily misled.

In summary, sometimes AI hallucinates and sometimes it does not. It’s very difficult to tell whether AI is telling the truth or fabricating answers, especially when you’re processing a huge load of information and do not have the time to individually double-check each source. Even paid models are not immune to making mistakes, especially when the user fires off multiple questions within the same conversation or uses chain-prompting techniques.

Which, by the way, is what most of us tend to do.

Fortunately, with the help of blockchain technology, we can reduce the chances of this happening.

‍

Integrating Blockchain Technology with AI Web Search

One of the best use cases of blockchain for AI-powered search would be to help maintain public databases of information that AI can refer to. For general web searches, storing all this data within the blockchain itself may be impractical since compared to a traditional database, data within a blockchain usually takes more time and effort to retrieve. The data is sequenced in blocks, and a scanner would either need the exact hash of a piece of information or enough time and processing power to sift through the entire chain of data.

However, a lightweight and efficient blockchain could work in tandem with a public database. The blockchain can act as an auditable ledger containing verifiable records of when and how new information gets added to the database, including the source of the information and its level of credibility. It can also maintain a list of credible sources and flagged content that AI programs can use to identify spam and misinformation. Search engines do have databases and algorithms in place to filter out harmful or inaccurate content, but they are not open to the public the way a decentralized blockchain is.

Blockchains are especially useful for tracing the origin and history of content. It can store hashes of original documents and create time-stamped records to help establish whether or not information has been modified, misinterpreted, or otherwise tampered with.

Blockchain technology can also be used to create reputation systems for websites and content creators. Currently, Google relies on a combination of trust signals to rank websites for expertise, authority, and trustworthiness, but with the exception of upvoting and downvoting results or hyper-optimizing a website’s SEO, the general public doesn’t have much of a say in that.

Yet another benefit is incentivizing users who curate high-quality content, flag spam, or identify correct sources by offering tokens to reward behavior that enhances the search ecosystem. This way, not only can the content itself be monetized, but also actions done to improve the content.

Another issue the blockchain can help solve is user privacy, intellectual property rights, and data protection — something AI models like Google’s and OpenAI’s have been accused of violating. Using smart contracts to manage data sharing preferences and zero-knowledge proofs to protect user data from unauthorized access can be one of the biggest benefits blockchain technology has to offer.

‍

Real-World Applications Across Sectors

Education

Educational institutions and learners rely on web searches to access scholarly articles, course materials, and research data. Academic institutions can utilize blockchain to create a decentralized database of scholarly articles and educational materials. Blockchain technology can be used to verify author identities and cross-check the identity of content. This helps ensure that educational materials sourced through AI-powered searches are reliable and trustworthy. 

Entertainment

Blockchain's immutable ledger can also be used to protect intellectual property rights in the entertainment industry. By registering content on the blockchain, creators can assert ownership, making it easier for AI-driven search platforms to identify and filter out pirated content. This not only supports creators' rights but also ensures that consumers access content legally and ethically. 

Health

In no other area is the battle against misinformation as crucial as in the health sector. People turning to the internet for advice on vital topics such as health, finance, and law— a category often dubbed Your Money or Your Life (YMOYL)— risk serious harm from wrong information. Yet as we all know, much of the “expert” content online is written by people without sufficient knowledge of these topics.

The best users can do is to stick only to credible websites or to look up the qualifications of the authors either through their author bios or their social media accounts. A blockchain-based reputation system that holds immutable records of author credentials and work experience would be a reliable way to ensure the trustworthiness of the sources.

Business

Enterprise blockchains can be used to create a decentralized database within an organization. AI-powered search can then enable employees to access this collective knowledge base easily. For example, an employee looking for information on a past project or specific expertise within the company can use an AI-powered search tool to quickly locate relevant information or the right contact person. 

Already, companies like Amazon Web Services are using blockchain services to create and maintain credible decentralized databases that can be accessed by the public. The blockchain acts as a comprehensive ledger, documenting database modifications, source material, instances of regular fact verification, timestamps, and more. This is more credible than traditional databases which simply present data without any historical record or evidence of authenticity.

Integrating blockchain technology with AI-powered web search addresses many of the challenges faced when searching for information online. The synergy between the two can potentially create a more reliable and transparent ecosystem for web search. This ensures that each time we use our search engines — or search chatbots — we get information we can trust.

In the case of Bitskwela’s CFO, we could have a credible public dataset containing the names of crypto startup founders, including Bitskwela’s. A blockchain would manage the record of all information added to the dataset. We could also have a blockchain-based trust ranking system that rates the official Bitskwela site and news articles as verified and high-quality sources of official information. Random tweets or social media posts, especially by accounts not officially associated with the company, could be ranked as low-quality sources.

This would help direct search engines to the most credible information available without having to manually instruct AI how to do its job. That way, we could avoid inaccurate results and go from this:

To this:

Of course, despite its potential for database management, simply integrating blockchain technology doesn't automatically eliminate misinformation. Search result algorithms can still misinterpret information or introduce bias, and as the nature of content evolves into videos, 3D simulations, and games, we’ll need new methods to identify and eliminate low-quality content. Likely, it will be a mix of both blockchain and artificial intelligence working together in synergy. 

However it goes, let’s not forget to always do our own research (DYOR) and never blindly trust anything online without first verifying the facts.

‍