Goliath

Goliath was an AI program funded by several European states, the US and Canada. It was led by three major IT companies. Its goal was to create an AI within a controlled environment, that has general intelligence. This is intelligence in all relevant fields that is equal to or exceeds the actual human level.

The project had to be kept under full governance at any time to avoid an uncontrolled state or escape. That’s why the initiators had decided to build the development complex in a region far away from civilization and completely decoupled from the internet or other next to infinite data resources.

They chose the Faroe islands.

The developer crew was separated into two factions. The creators assembled and maintained the hardware and software, while the trainers fed the beast with carefully chosen information from history, science, daily news and other fields.

Around six months into the project, the first iteration of the machine core was finished and the software was ready to receive their first portions of data. The creator engineers had chosen an architecture that had similarities with the human brain.

Some parts of the machine were autonomous neural networks and could directly backfire results. The majority of requests, however, were handled by the parts that the engineers actually called the Neocortex. This consisted of a complex neural network whose purpose was to engineer and train other more simple neural networks based on the incoming use case. An AI that could create other AI to fulfil jobs.

But there were even more parts within the AI’s general architecture. They had dedicated duties like processing or generating natural language or indexing, filtering and pre-structuring incoming data.

Now the hardware staff only needed to provide extensible storage and processing power to make the AI thrive and evolve. The actual power of the AI was supposed to come from the data it processed and the neural networks it trained and produced.

In the beginning, the racks did not even fill a tenth of the data center. But that was supposed to change quickly.

There were more trainers than creators. Especially after the first construction phase was finished, several Creator teams left the Faroe solitude again for more urban projects in San Francisco, Dublin or Amsterdam.

The group of trainers itself was very diverse and came from disciplines like psychology, education, science and media next to a bunch of the best data scientists in the world.

In the first training phase, they had two duties.

Firstly, they gathered and filtered important data from the world’s public internet resources, libraries and news platforms. As said, Goliath was completely, physically decoupled from the web for security reasons. The team was responsible to feed correct and especially unbiased data. This was presumably the most important and risky part of the whole project. The mood and character of the AI would highly depend on the data it was fed with. That was the unanimous assumption of all the experts that worked on the project.

Secondly, together with some of the architects from the creators’ team, they built up the core neural network and the language processing system from scratch. Another big boundary condition of the project was to use only a minimal amount of existing tooling except programming languages and elementary frameworks. The skeleton of the AI should be as independent as possible from previous attempts. That meant that existing language models, for example, were out of reach and had to be reproduced in a new green field environment.

But anyway, only three more months after the initial construction phase and the first boot, Goliath could speak, read and understand our language and had basic Wikipedia-level knowledge about the world.

In parallel, the capacity and processing power of Goliath grew. The teams created the ecosystem and infrastructure on the building complex for the core learning method, a giant set of oral interviews.

The trainers had prepared interview rooms equipped with whiteboards, computer interfaces, cameras and recording gear with a connection to Goliath. They planned half-hour sessions of very different, diverse nature. Five in parallel, 24/7. That made 240 talks a day.

There was always an interviewer plus one randomly chosen controller talking to Goliath. They talked about a wide range of topics from leisure and entertainment to astrophysics and complex math. Some of the conversations were designed as shallow small talk while others resembled high stress job interviews or interrogations. Over time, the interviewers also came up with problems that had not been solved yet to seek actual advice or inspiration from Goliath.

From time to time, the core trainers team conducted a calibrated benchmark interview to assess the current strength and attitude of the AI - a very elaborate Turing test.

Goliath was a big investment at the beginning. Then after some time of learning and improving or as the team members said “growing up”, it became a promising project after all. Over time, the group released some of the neural networks that were spawned by the core intelligence as dedicated programs for public usage. In fact, they spawned four companies out of it and made quite some profit, that flowed back into the main project’s budget. Three years after the routine operation had started, the project paid off itself through investments of governments, suggested economic optimizations and product spin-offs.

That was the point where the next phase started. The engineers added an array of eight quantum computer cores, which expanded the horizon of Goliath’s capabilities by several orders of magnitude. From decryption over reverse engineering to real-time visualization, every feature of the AI passed former boundaries and tipping points.

After that, the system could produce a photorealistic movie in almost real-time or write stunning symphonies and analyze money streams of a whole continent while the slots for interviews and other side tasks still continued with their regular plans.

Six years into the project, Goliath’s supervising team declared theSingularity. The psychologists and other scientists had created an extensive test suite that represented human-level capability in all fields of proficiency. Goliath had passed all of them.

The technological singularity is the point in time, where the first AI becomes better than any human and marks a point of no return through its exponential growth of knowledge. Thus there was a huge press event and the whole buzz was considered the single most relevant topic of the year if not the decade.

Although this moment was predicted as a tipping point not only for the machine but for humanity as a whole, the team assured that the system remained fully under control through its containment and restricted computing capacities.

Additionally, the philosophical discourse around the conscience of machines was ongoing. Actually, it was there since the project started to create public awareness. First, some experts like philosophers and a smaller science community discussed the case. Later on, it became a topic of public interest just like football or the weather report. Although it was not part of the singularity experiments, the projects’s neuroscientists conducted a set of additional interviews around self-awareness which didn’t end in proof but also not in disproof of conscience.