I have a vision.
A vision where my digital assistant’s only task is to understand me, to support me, like a head butler in the household. Where it is in control of an army of specialist AI agents, each with its own task to make my life easier. To me this is the true digital happiness: innovation to make my life easier.
Chatbots are not this development.
In my opinion, there are a couple of serious flaws in the approach that chatbots have taken:
- Chatbots thrive on learning about us, data is important. After the hype, if we think it over, I wouldn’t want to share my details with every single chatbot out there just like I don’t want to share my details with every individual on the street.
- Regarding the mining of data to improve the service, this in itself is quite good. However, we don’t need chatbots for this, the same data can be collected with any frontend (even more effectively since navigation is easier than language processing).
- The vast majority of chatbots out there are a simple reimplementation of that existing frontend into a textual frontend. The latter is more obscure, harder to navigate, and often of poor quality. The minority out there that work well have spent significant development time into it, but that wouldn’t scale.
The one innovation that first seemed like a nifty idea was: slow down the conversation to make it look human.
I’ll repeat it:
Slow down the conversation… to make it look human…
If there was ever a sign chatbots are not the best fit for AI, I think we found it.
For getting to that next phase into true application of AI – not data mining but actual intelligent behavior – I envision the following steps:
- Digital Assistant specialize in getting to know me. They’re currently well under way and will soon realize it’s just not feasible implementing every task within the assistant itself. Execution of tasks should be delegated.
- Where in the past we implemented all business logic for applications and services ourselves, this is where AI will play a huge role. Specially trained agents for single tasks (a.k.a. Specialized AI) will be spawned when a task needs to be executed.
- To be free to execute a task independently, an agent needs to be able to traverse the internet autonomously. A platform infrastructure will arise where agents can move around, consume services offered and interact. The internet is no longer a network of communication lines, but a world inhabited by agents.
- An agent needs two resources we bring with us every day: an identity token and money. And since we don’t know where they should be able to go, it should be in a system that’s present everywhere and inherently trusted: enter the Blockchain.
Since Federated Identity is simply not sufficient when we don’t know where trust relations are needed (and we can’t just trust everything), the next generation of identity management should be blockchain-based. An agent should bring a delegated identity token with claims describing its owner and its creator at a minimum and be extendable so it can collect authorization claims to specialized services.
- Lastly, being mobile means an agent should be able to run on many different platforms. An interpreted language seems the most natural candidate. Luckily if this comes to pass the need for graphical frontends will diminish, leaving a vast army of frontend developers free to start implementing AI libraries and agents, supporting all tasks you desire.
I have a vision.
An open world where AI systems can be free to roam around, assisting us in reaching new levels of digital happiness.
Let’s make this happen.