Orion is gaining a foothold by representing and activating corporate knowledge, but its goals are greater.
Humans have two significant weaknesses – they can handle no more than four things at once at a conscious level, and while they can learn quickly, they are very poor at “unlearning” things, essentially having to die off if what was learned is wrong and deep (ulcers was a good example – many doctors found it difficult to come to terms with the fact that bacteria can cause ulcers – they refused to believe it). The technology we are proposing has the same problem in unlearning things (for the same reason – making new connections where none exist is easy, severing connections which support later learning is hard), but it does have a Kill switch, so that all or selected parts can be relearnt.
For hundreds of thousands of years, very little unlearning was required from one lifetime to the next. With our rapid increase in the use of energy has come a much greater stress on our environment, with significant changes apparent in a person’s lifetime. To handle this rapid rate of change, we will need to make decisions that depend on more and more simultaneously interacting inputs, to the point where the very low number of things we can handle simultaneously becomes inadequate. An obvious difficulty of ceding decision-making to a machine is that the machine will not be able to explain to us why it made a particular decision – we would have to hold too many things in our heads simultaneously to understand the reasoning.
Alien environments are the most drastic example – since before we are born, we are swaddled in an Earth environment – first gravity, and then later the atmosphere, and how plants and animals, including bacteria, behave in that environment – everything we learn is coloured by our environment. We can’t expect humans to have their memories wiped and relearn everything in an alien environment, but we can make machines do this.
The astronauts who went to the moon were loud in claiming that we would need vital machine support to go to Mars – admittedly the computers of the 1960s were primitive, but the communication delays mean the machine itself has to be in the alien environment, with immediate response. Most of the knowledge it will use to build its internal structures will be in textual form – the laws of physics won’t have changed, but the environmental colouring of them will have. While we will create an Earth-like environment, disaster will lurk at the periphery in every dimension.
The machine will have to stop us metaphorically driving on the Earth side of the road when we get tired.
This is an area where Big Data is No Data – how many disasters on Mars should we plan for, until we have the data to avoid them? The obvious answer is none. Big Data can only work where very little changes, or it continues to use data after significant system change has occurred, making Big Bad Data a more accurate moniker.
Is Orion up to the task?
It was begun as a decision support system, to avoid the primitive decision methods used in spreadsheets and other tools. The decision path is dynamically modifiable (that is, not programmed), so it can use what it learnt as a structural change a year ago or a minute ago (the approach is nothing like that used in a self-driving car – Orion is an active structure, where activity takes place in a structure continuously being modified by words among other things, while a self-driving car is using an algorithm, in an environment which completely lacks the fluidity of language – a few words can open vast realms of knowledge, while turning a corner leads to more of the same).
Orion is built to give high reliability, so it is easy to see where it has made a mistake, and we don’t need to accept the high error rate of a probabilistic approach – being right 80% of the time (or even 95% of the time) is disastrous in a long-term high-risk environment.