Why should we, as a species, seek to create human-level artificial intelligence? To AI researchers and engineers, imagination and curiosity could be enough justification.

But to everyone else, a more substantial answer is required. What is our purpose and benefit in doing so? What is the future we are trying to achieve with AI?

Too many pursue this subject without a clear vision of what they are building, or the consequences. As a result, observers both hope and worry, torn between the promises of science fiction and the risks of uncontrollable AI — or uncontrollable corporate greed, powered by AI.

Uncertainty about the future of AI is what creates fear against it. If AI is to be beneficial and trusted, we must design it accordingly — and good design requires a convincing vision of what we are trying to build.

This blog will propose an alternate vision for AI, one that is based on automating the flow of knowledge in our society, rather than automating work.

We will explore what knowledge is, and how knowledge works, in an effort to design that future. We will discuss the consequences and applications across many facets of our society — in our relation to education, technology, healthcare, governance or even travel and transportation.

But today, let’s start with how and why we should create this future of knowledge automation. This will be a long article, but a necessary introduction to this blog.

Disclaimer: to keep things simple for now, I will rely on the common understanding of the word “knowledge”. A more advanced discussion of the nature and mechanisms of knowledge will be the subject of a separate article.

To perform the smallest action as an individual, or pursue the most elaborate enterprise as a society, a complex assembly of knowledge is required.

We think, speak, and act thanks to knowledge. It powers our intelligence. We learn, teach, engineer, cure, judge and debate thanks to knowledge. It powers the progress of our society. There is knowledge that we combine and create within our own minds, and knowledge that we exchange, build together.

Without knowledge, there can be no real intelligence. We could not recall how to tie our shoe laces, we could not agree on the meaning of words. We could not even disagree with each other, for there can be no communication without knowledge to communicate.

Our entire civilization is built on knowledge. All of our progress, be it cultural, technological or otherwise, was made possible by our ability to build new knowledge upon the knowledge of previous generations. We are, inescapably, a species and a civilization of knowledge unlike any other we know of. A species that relies on knowledge, consumes knowledge, produces knowledge, and exchanges knowledge at an ever-increasing pace. If we can optimize the flow of this knowledge, we can optimize the progress of our civilization.

To do so, we devised ways of exchanging knowledge at ever greater scales of time and distance. We made it possible for the words of a teacher to reach the minds of students born two millennia later. We made it possible to spread information across the world in seconds. Teachers, books, newspapers, post offices, libraries, universities, telecommunication, search engines and social media all contributed to increase the flow of knowledge between individuals on earth, and constitute our knowledge infrastructure.** A stop sign, your doctor and even the back of your cereal box are part of this knowledge infrastructure, bringing to you knowledge you need or knowledge you seek.

While we may want to call it an information infrastructure instead, ultimately all information and data we record and exchange is meant to reach a person. Be it a road sign, a spoken sentence, or a number on a computer — all information recorded in our infrastructure exists to eventually create knowledge in a human mind. Knowledge is never really transferred, but recreated, as we observe and interpret this information.

Information is the medium of exchange, but knowledge is the goal.

Every invention that reinforced our infrastructure overcame one or several barriers, that prevented knowledge from flowing more efficiently. This may be a barrier of language, distance, time, cost; it may be the lack or unavailability of knowledge, or even an inefficient interface to access it.

Ultimately, anything that stands between us and the knowledge we need can be considered a knowledge barrier, even if it is knowledge yet to be recorded, or knowledge we don’t know that we need.

The latest revolution of our knowledge infrastructure was the internet, which broke barriers of distance, speed, availability and universality of knowledge access. It allows information to travel nearly instantly from one end of the world to another, or lay dormant in a cloud. This mesh of globally interconnected computers was a true revolution when it comes to knowledge infrastructure, and intuitively it‘s hard to think we could go much farther than that.

But even though the revolution and its impact on our societies and economies were massive, even though major barriers were spectacularly broken, that does not mean the job is done. By far.

While internet provides access to massive amounts of information,

• It takes plenty of time to find knowledge, search for it. The more complex a question, the harder it is to find answers.

• Not all knowledge can be found on the internet, or found in a way that’s easy to understand. Sometimes you do need to contact a repairman or consult a doctor, despite your best efforts.

• Information is recorded in many different languages, so we can’t all access and read everything as easily. The languages we understand dictate which subset of the world’s online knowledge we get access to.

• Not all regions are equally connected; and not all regions record equal amounts of information on the internet. Therefore, local knowledge availability and accessibility may vary.

• There is no agreed upon organization of knowledge on the internet, everyone dumps information where they please. Knowledge is fragmented across many sources and forms, for the convenience of adding information to the internet, but at the inconvenience of all users attempting to navigate through it.

• Internet is full of opinions, influences, ads, even manipulations and scams competing for attention — it’s no easy task to navigate, to isolate the truth and the source of quality knowledge from the rest.

• The algorithms that are meant to help us navigate this storm are partly motivated by ad revenue. The more you spend time searching, the more ad revenue you generate. The less helpful and accurate the search results, the more time you spend searching. Strangely enough, search engine results have only become less accurate over the years.

• But this may also be due to the increasing amount of information humanity generates online on a daily basis. The more information we generate, the harder it is to search for needles (precise, specific knowledge) in the global haystack.

• The interfaces we use to navigate the web expose us to many ads, popups and GDPR nonsense, that we would be better off without. Of course, it does help pay for freedom of access; but paradoxically, it also prevents smooth and efficient access to information.

In short, we still spend too much time and effort searching the internet to access knowledge, and it is time and effort that is sucked away from our lives. Yet we know it’s possible to access knowledge with ease: ask a knowledgeable person. They will answer right away.

Ultimately, the critical flaw of our internet-based infrastructure (and its interfaces) is that it does not understand us, nor the knowledge it contains.

That flaw is nothing new, and it is shared by almost every knowledge system that preceded the internet, such as books and libraries — but not with people-based systems, of course. Lack of understanding is at the core of almost every problem we’ve listed above, and hints at the most effective solution to resolve them.

If our knowledge infrastructure could understand, then it would be able to locate specific knowledge by meaning, like we do. Regardless of language, knowledge could be located and shared by what it is, rather than its http address. We could detect and filter misinformation and scams at the source. We could create a cleaner, more efficient, more accessible internet.

But the internet we have today was built to share text and media at addresses, not to understand and answer questions. Search engines may attempt to solve this by directing us to relevant addresses, but they often fail to do so effectively. They do not understand our queries and their nuances. They do not understand the content of the webpages they return. But you do.

When you use a search engine, it’s only doing half of the job. You must yourself search through the dozens of websites the engine returned, because only your human mind is capable of understanding the content. Most computer algorithms are unable to replicate that understanding. You are the real search engine.

It always takes human minds and human time to extract, refine and wield knowledge. It always takes a human mind to jump around from interface to interface, database to database, device to device, book to book. So far, only human intelligence is capable of knowledge intelligence, and only knowledge intelligence is capable of hunting down knowledge within information. We are the pillar, and the bottleneck of our own knowledge infrastructure.

If we wish to build the next generation of knowledge infrastructures, we need to find a way to remove ourselves as the bottleneck. We need to build a knowledge infrastructure that understands knowledge on its own.

This is where AI comes in.

In order to build this future, there is no other choice: we need to create machines capable of replicating our own **knowledge intelligence **— the part of our intelligence that allows us to learn and make use of any knowledge. No other type of AI will do.

As we observe, hear, touch, smell, we are able to extract knowledge from our senses. We are able to memorize, access and transform that knowledge in order to create new knowledge. We are able to predict, make plans and take action using knowledge. There is a logical process to this, one we have not yet fully elucidated. That is what I call our knowledge intelligence.

In a sense, knowledge intelligence is a subset of what our human intelligence is capable of. But the knowledge intelligence we are capable of is also limited by our biology. We are flawed vessels of knowledge — we did not evolve to have perfect recall, nor did we evolve to be unbiased and incorruptible.

Machines, on the other hand, will be what we make of them.

Machines offer the opportunity, if we are able to understand the inner workings of knowledge intelligence, to replicate that intelligence and achieve a greater form of it. An AI that is capable of understanding the same knowledge and performing the same knowledge operations as we are, but with the capacity, reliability and speed of computers.

I call this knowledge-centric AI an Artificial Knowledge Intelligence — in short, an AKI.

This new concept may seem unnecessary, when more general terms such as human-level intelligence (HLAI), artificial general intelligence (AGI) and artificial super-intelligence (ASI) are already widespread and capturing the imagination of many. But the concept of AKI differs in important ways that can change the trajectory of research, and the potential effect of this AI on society.

As opposed to other forms of AGI, an AKI cannot simply rely on an unreadable network of mathematical values to make decisions. It’s not enough for it to appear to be performing intelligently most of the time — it must be able to fully explain every single one of its choices. A knowledge-based AI must be able to justify every decision with knowledge, and justify every piece of knowledge with its source.

Consider ChatGPT, for example. The chatbot, recently released by OpenAI, is not only capable of speaking several languages, it’s also able to share with its users the extensive knowledge it has acquired during training. However, all that knowledge has been compressed into a network of numbers. All information about sources is gone. It’s impossible to look inside its memory and confirm what the AI does or does not know. The AI itself doesn’t know — it’s unable to prove its claims by quoting sources, and unable to understand when or why it is wrong. Even if it could, there would still exist the possibility of mistakes or manipulations, so long as we cannot “check inside”.

The better form of AKI is one whose internal structure of knowledge can be transparently explored, and verified by developers, users and independent watchdogs. It must possess a readable, verifiable structure of knowledge by design — so that we can ensure it is acting based on truth, evidence and logic at all times. So that we know it is making the right conclusions for the right reasons, not with false information or mistaken assumptions.

An AKI is the only form of human-level AI that could be transparent enough as to be verifiably trustworthy, and verifiably beneficial to mankind. Just as we verify open source software, we should be able to verify knowledge-based AI.

And as opposed to AGI or ASI, a successful AKI does not need the capacity to do anything and everything that we can. It does not need to drive, paint or play videogames better than humans. It is only meant to exchange knowledge with us — to assist us, not replace or surpass us.

This has implications in how the AI could be perceived. An AKI limited to the purpose of assistance would be much less of a threat to our society, while being of great benefit to it. Governments could even ban the use of AI to generate profits — yet still, knowledge would flow, and AKIs could continue helping people learn, and wield greater knowledge, for their own benefit.

How do we integrate knowledge AI into everyday life?

I don’t envision AKIs as invisible, all-knowing systems, working within the shadows of the internet. Very much in line with sci-fi and current AI assistants, I see them as more limited, but more personal knowledge agents that act on behalf of individuals and organizations.

The knowledge agent would be a hyperqualified assistant, and our interface with the world’s knowledge. It could help us collect, protect and organize our knowledge, both private and public.

Whenever we seek knowledge, it will understand our context, it will know where to look, it will automate the search until it finds answers. When necessary, it will access and assimilate the collected knowledge of professionals. It may become a doctor one day, and a lawyer the next. When appropriate, it may make new knowledge available to our public knowledge infrastructure, thus contributing to growing humanity’s knowledge, and keeping it up to date.

It will help us learn, it will help us repair, it will help us diagnose, it will even help us know our rights — and defend them. It will help us navigate foreign countries and cultures, help us purchase the right products. It will be an assistant, a teacher, a mediator, a guide, an advisor of anything. With the right network behind it, it may leverage the knowledge of millions of experts, patients, users and situations, to provide us with the best quality of advice regardless of the subject.

A knowledge agent may one day run from a home computer, operating on behalf of one person or a family, using our less powerful mobile devices as proxies to interface with us. It may belong to a corporation, hosted on its servers, smoothing the flow of knowledge between teams, coworkers and clients. It may belong to a modernized hospital or clinic, to improve the coordination between patients, nurses, doctors and labs. It may belong in a small business, taking calls and appointments, or answering questions from customers.

Many such examples will be discussed in future articles. A large part of this blog’s purpose will be to explore the place knowledge agents may take within our lives, our businesses, our institutions. It’s not good enough to create an AKI or AGI and let it loose without a plan. We should know what we want to achieve and how, in order to achieve it well.

At first, we could imagine our agents using the internet in a similar way as we do. They may use our text-based search engines, or memorize the best websites or applications to find specific information. Our agents may be expected to read websites, e-books, scientific papers, and consult each other. They would gather information, compare it, summarize it in similar ways we do it. They may directly make use of databases and APIs as programmers would. Thanks to their relentless digital minds, they could automate a search task of hours down to minutes.

But why settle for minutes when we could do the same in seconds? To make the next leap in knowledge automation, we would need to change the way information is organized and discovered on the internet.

Knowledge intelligence opens up the possibility of memorizing and searching by universal meaning, rather than language-specific text. In that situation, why continue to index websites by keywords, and why continue to use inefficient text-based search engines? We could directly register new information with AKIs, and let them organize the structure of that knowledge on the internet. This may result in the obsolescence of search engines as we know them.

We can expect the need for websites to continue as long as people choose to express themselves in that form. But for a company launching a new product, for a shop changing its opening hours, or for a government changing its travel requirements, the most effective way to dispatch information to the world may become their official knowledge agent, rather than their official website.

The decisive factors that will drive adoption of knowledge agents will be how easily and how fast we are able to share or access knowledge with them. Despite its flaws, ChatGPT gives us a preview of that future. It shows that an interface capable of properly understanding and exchanging knowledge with its users can be far more convenient to people than dealing with search engines and websites.

But how should AKIs structure and spread their knowledge? Since no AKI can know everything, where should knowledge be stored? And as knowledge is passed around, how can we ensure its authenticity?

The answers will depend on who designs and who owns the technology. Whoever creates AKIs, and whoever controls their knowledge networks, will have the fate of our knowledge infrastructure in their hands.

A major corporation may have little regard for the idea that humanity’s knowledge should belong to all of us, and their approach to the technology will differ accordingly. They will likely choose to centralize all information on their servers, so they can sell ads or a subscription in exchange for access to the world’s knowledge. They may choose to lock your AKI to use only their knowledge network, or their partners’, so that they can sell you their products — or even control your knowledge on behalf of their state.

Whoever controls your access to knowledge will be capable of dictating their own truths, whether they intend to or not. As long as one entity has control over our knowledge, they present a single point of failure that can be exploited by hackers, corrupt governments, or even the next CEO. It’s a dangerous direction to pursue, yet in the world we live in, it is a very likely one.

We need an alternative — a more decentralized approach in which no single entity can control our knowledge infrastructure. We might see AKIs form an independent network of agents, each with knowledge capacity and bandwidth to share. Each with their own specialized knowledge, but also aware of who holds the best knowledge for what topic.

All the world’s knowledge may be spread across those machines, and companies or users with additional storage space to share may also contribute to the network’s total capacity. Knowledge may be duplicated or triplicated across the world, in an effort to protect it from catastrophe, war, and undue influence.

In this decentralized approach, the network must be carefully designed with rules to protect it from manipulation at all levels, and to guarantee the veracity of information. Whether it happens with a human lie or a modification in an AKI’s knowledge structure, we cannot allow false information in a single AKI to propagate to the entire network.

As a government makes an official declaration, as a public figure makes an statement, or as a business recalls a product, we must always authenticate and remember the source. As knowledge is passed around, duplicated, reported, we must ensure the content and the facts of its origin remain untampered with. When bad actors release false information, we must be able to follow the trail back to its source, and protect the network against those who seek to corrupt it.

Building such a decentralized network of knowledge may well be the only way for us to not only build an infrastructure capable of exchanging the world’s knowledge, but also one that can be trusted. In our knowledge civilization, knowledge is too often distorted, fabricated — weaponized — to influence public opinion or consumer behavior. To sell cigarettes, win elections, start special operations, or pretend climate change doesn’t exist.

If the future of knowledge does lie with AI capable of knowledge intelligence, and a network designed to trace and verify knowledge, we may be able to change course.

But it will be for nothing if their integrity is compromised. To ensure the best future, AI must be reliable and transparent in its operations; ideally created for the common good, rather than for corporate interests. Networks of knowledge must be created that are resistant to misinformation and manipulation; ideally owned and controlled by everyone — or no one.

A universal network of knowledge cannot exist without independence and objectivity.

AIs have so far failed to meet our expectations of intelligence, not because they lack intelligent abilities, but because they lack the specific ability to understand knowledge, and exchange it with us reliably.

With knowledge AI, we could automate not only the search for knowledge, but the organization of it. We could form the most complete collection of the world’s knowledge, and make it easy for anyone to query it, and learn from it, regardless of age, country, language or proficiency with technology. Knowledge agents could become the optimal interface to exchange knowledge — an interface based on human knowledge and human-like interactions, rather than computer interactions.

We are a species of knowledge, whose intelligence and history is built on knowledge. Knowledge is our education, our skills, and our ability to succeed. It’s our individual and collective potential.

It’s not with the automation of work that AI will provide us with the greatest benefits, but with the automation of knowledge. Knowledge AI is the future of our knowledge civilization.

You’ve made it to the end! Thank you for reading.

In the next articles, we will discuss the barriers that create friction and inequalities in our access to knowledge; We will look at what knowledge agents could look like, and provide practical examples of their effects in our lives; We will investigate the mechanisms of knowledge in our minds, and try to figure out if knowledge is really what we assume it to be.

Stay tuned!