Effect.AI: the Hybrid Automation Pipeline
Effect.AI is elevating AI to a new practical level with a man-machine combination called The Effect Network (TEN). Combining weak AI systems with strong human intelligence components is a game changer for business automation. Effect.AI is putting this to work by introducing the first autonomous application on top of TEN, named Rosette.AI. Rosette is a unique translation engine and touches the core of human intellect: language.
This article will shine a new light on The Effect Network and will demonstrate its use as a practical AI creation platform. I’ll first take you back through a bit of history to clear the water around what kind of AI is being built today. Then we’ll look at TEN as a framework for applications that require a combination of machine and human intelligence and showcase it with a new product called Rosette.
Introduction
If you’ve followed Effect you know that we’re building a Network for Artificial Intelligence. Over the last few decades, Artificial Intelligence has become a very broad and widely used term, with the associated hype.
The purpose of any practical AI is simple: it should save us time and energy. That’s it. The one and only goal. We’re not talking about “Artificial General Intelligence” here, like Skynet or GRTA. I’m talking about simple and solid AI that beats you at chess when you don’t have a friend around, that guides you through an unknown city and does your shopping on a voice command.
“weak is winning”
Between the movies and books, the politicians and the over hyped startups, an image is projected of AI as a dominating force that will take-over. It can be hard to separate the hype from reality. A practical look on AI is that they’re just computer programs that make our lives more efficient and fun and this is the view we’ll be embracing going further.
As early as 1980 scientists and philosophers were discussing the concept of strong AI and weak AI. This illustrates the difference well:
- strong view: an AI system can think and have a mind
- weak view: an AI system can only act like it thinks and has a mind
Strong AI is the AGI from the movies, the humanoid. The holy grail. It’s paraded on the hype train and is often misunderstood for what companies are developing today. Now, weak AI is more like the intelligent systems that we’re accustomed to: from lookup formulas in Excel to the artificial neural networks in your Nikon camera.
“AI as a language”
When we talk about a toolkit for practical AI development, we’re talking about building out and connecting systems that currently exist. These systems are everywhere: inside Tesla’s Autopilot and powering Apple’s Siri. They’re making our lives more efficient and give us bearing in the digital age.
Ready for some Geek history? Just 49 years ago, before the time of floppy discs, people were building computer algorithms by punching holes into plastic cards. These cards were called “punch cards”. A machine would read in such a card and know what to compute by detecting the drilled “holes”, similar to computer bits. Looking at the technology that we have today this was a very primitive way of building algorithms. The means of programming have evolved rapidly over the years, with the introduction of many new hardware and software systems. To date software development is getting more efficient with the advent of smart editors, IDEs, frameworks, SDKs, you name it.
“The Effect Network”
In this perspective TEN can be seen as a new abstraction for building complex algorithms. Its axioms are data that is flowing in from a source, like an invoice PDF or camera picture. The means of combination are the algorithms; they perform transformations on the data. And the means of abstraction are the combinations of algorithms that can be used as computational units themselves.
The Effect Network solves the complex issues that arise when running and combining algorithms in the wild. The network is fueled by the EFX token, just like cars use gasoline and humans need food and water. TEN is like a self-sustaining city that never sleeps.
Alright! That’s enough theory. It’s time to go down the rabbit hole and show you what this all actually does.
The Man Machine
The potential of TEN is best described with an example. Below is the setup of a company that integrated its invoice processing pipeline with the closed alpha version of the SDK. This workflow revolves around catching incoming emails that contain a PDF invoice as an attachment. The invoices can be in different languages but should all end up in the same accounting software. Also, some of the invoices contain sensitive information like banking details that can’t be disclosed to the public. These invoices are first checked by a confidential workforce under NDA that will anonymize the data. By combining a human workforce, an email server and 2 simple algorithms, this workflow is completely automated.
On their own, these steps are simple tasks that are readily available. Combined they create a unique pipeline of automation. The combinations of human intelligence and algorithms within TEN is vast. It acts almost like a new type of automation language to develop upon and experiment with.
So what’s available on TEN?
Next to a wide selection of simple open-source algorithms and data collection units — such as email servers, FTP watchers, and HTTP endpoints — any campaign on Effect Force is available as a computational unit. There is no limit to what you can process with this protocol as long as latency is not crucial. The table below shows a small grab of the options.
In TEN, each of these components is its own independent entity. This means that the “cost” of the pipeline is the sum of its components and this “value” is directly distributed to them. An algorithm is its own entity (could be an AI DAO) that pays for its computation and has a kickback to its creator, while the Human Workforce is actually paying labelers around the world as well as hiring a campaign manager that guarantees throughput and quality.
This combination of man and machine is at the core of the practical power of TEN.
Rosette.AI
We have described how TEN is a framework that can be used to build practical applications on top of algorithms. At Effect.AI we’re applying this to help clients process and learn from their text and voice data by implementing automation pipelines, like the invoice integration described above. This technology has now matured far enough that we are introducing a new concept: the development of applications. These are processing pipelines connected to a user interface, allowing direct consumer interaction.
To solidify our NLP efforts and to further demonstrate the capabilities of TEN, we are introducing a new project called Rosette.AI: the first application to be built upon and run exclusively on The Effect Network. Rosette is a translation engine that gives any entity, worldwide, direct access to a wide range of languages. Rosette outperforms existing state-of-the-art machine translation by utilizing the combined power of AI and Human Intelligence.
