Many countries explore ways to get not disconnected from the AI development even though the risks and opportunities of AI are highly controversial. This is the reason why we started to develop the AI Leadership Framework. Further helping hands are very welcome.
Since we are working in the Artificial Intelligence space for quite a while, we decided to put up an initial framework for country leaders.
AI Leadership Framework
The “AI leadership framework” is a project in development. We are working with a large number of great minds from over 20 countries to postulate a framework that can be used as is, or as a base for a country’s own AI Leadership Strategy.
1) Data Awareness
Data is the new row material for the highly developed world. What was Gold, Coal, Oil and so forth, even healthy soil, in the past is data in the future. There is an abundance of data. The key is to harvest the available data. Being aware of the fact, that we all are creating new data, billions of data every single day, is the first step to create policies, infrastructure and motivation to share those date. Like in previous technology developments, the US has been in a leadership position early on. Data power houses like Google, Amazon, Facebook, the credit card organizations and telecommunication companies, already sit on data like no other country.
2) Access to data
The more data can be assessed the better the results from AI systems. In particular in the early days of any AI development, it is critical to get access to an enormous amount of data. Enormous means several hundred thousand records about the same topic. Over regulated and fearful societies and those with fear driven privacy campaigns are clearly in a disadvantages position. Amazon has access to trillions of dollars worth of shopping from past years. Together with data usage from their AWS business unit Amazon is one of the top player globally. Also Google sits on quadrillion of data points from all the searches and knows what people look for, why, what problems they have, what products they search and so forth. Facebook, unlike the other two has much more social interaction related data. The recent collaboration with credit card institutions provide Facebook a hot mix of social and commercial data and relations.
AI is no longer an IT discipline. AI requires top level mathematical knowledge, understanding of neural networks, understanding of reasoning and decision making processes, understanding in cognitive behavior, awareness know how, and the obvious image and speech recognition skills are a given. The vast majority looks at AI as a software code that is written by humans and can do at best as well as the programmer. That an AI system can not only perform computational processes but also build intelligent correlations, make then decisions and outperform humans in almost every predictable project is foreign to most of the postulation. The algorithms (mathematical procedures) allow AI systems to learn from hundreds of thousands of situations, then construct millions of situations similar to those they learned from and have a richer set of “experience” than any human ever being able to accumulate. While this is good – we need to know it.
4) General Education
In past major development shifts, the education of what’s happening was more or less based on the effort of the businesses who produced and marketed those new technologies. However the possible impact is too significant to only hope the education will work out. If 50% of jobs will be irreplaceable eliminated, we need concepts and education how who this will not turn out as a catastrophe. We need to explore options in advance and educate those who have been educated to look for work and do what others tell them to do. what opportunities are out there in this new world. For the past 300,000 years humans have been taking care of themselves, have been rather autonomous and been creative to survive. In the past 200 years that has changed. The main job was to look for work and just do it. We are in a way going back to more autonomy and self determination. While this is an amazing and positive development, for many people it is on open void they may not be able to fill by themselves. Education and guidance is a key aspect of leadership – always was.
5) Language Empowerment
The world’s information is by and large stored in English language. Even the use of the term ‘Artificial Intelligence’ is important to be held in English and not in a local language. One of the key success aspects of AI is the openness and willingness to cooperate globally – that means by default in English. Leadership by ‘closed shop’, and keeping everything close to ones chest and secretive will never lead to superiority.
6) Culture & Failure Tolerance
Whatever a team is creating, it is critical to get the early prototype rapidly into the market and work with massive amounts of actions. AI cannot be only tested in a lab – it needs to be tested in the market. The European or Asian perfectionism is counterproductive to AI development and will automatically make those actors fall behind. Google’s AI project that was used to learn about battery usage in Android smart phones moved within six months from concept to being used globally. We can learn only from errors. We will not learn from a perfectly working solutions. What sounds stupid to many engineers has been proven right over and over again. The motto Fail and fail fast is even more important in AI development.
Due to the competitive nature of any new technology, we most likely will never be able to find out who is working on what. Our brains are still the most private part of our lives – and hopefully will be forever. We only can and definitely should develop a work ethic that calls for voluntary transparency. Now, recent incidents demonstrated that a business may not trust its own government because of its sheer power. A government should not even attempt to play a controlling role in the transparency question. And since trust is one of the key issues in that, maybe a system like we know from the blockchain development maybe a future option. Another reason why we should keep governments out of this role is the lack of trust among governments themselves. The current suggestion is that AI scientists select a consortium of trusted, anonymous third parties who get tasked to oversee the development as such and analyses the potential risks – with no authority power.
8) Safety Policies
Another key aspect of the leadership framework is the existence of the “Safety Trio”:
1) Privacy & Data Policies
2) Criminal Acts Policy
3) Human Protection Policies
8.1) Privacy & Data policies
need to be established to not only protect the general population from misuse of data but also education about the consequences of not providing data. Data as such needs to be structured beyond the act of privacy protection but also the protection of data, that belong to the creator of those data. Yet, it needs to be understood that data can either belong to one person or legal entity in its entirety or belong to a whole group pf people or entities. The complexity and permutation of data needs to be carefully explored on the go.
8.2) Criminal Acts Policy
Most likely the act of hacking systems and stealing data, corrupting data or systems or destroying either data, systems or network infrastructure will want to be set as a serious criminal act. As systems get more sophisticated they also can cause more damage than in the past and can bring people and a whole country into serious trouble. AI leadership needs to demonstrate its sensitivity for protection while at the same time its openness for sharing data. As such, hacking needs to be elevated to a serious criminal act with substantial penalties. The criminals act policy should also expanded into political treaties where country leaders agree on respecting the criminal acts policies of each other.
8.3) Human Protection Policy
There is s great level of fear, that AI based Robots, so called autonomous machines, may be at risk of harming humans. Right or wrong, AI leadership requires responsiveness and offers a solution. A possible method is to legally enforce that every autonomous machine will need to have a mechanism to be remotely shut off by especially selected authorities and/or mechanisms. A structure like the domain name service where systems are distributed around the globe may also be applicable to distribute the shut of mechanism. Robots without such a mechanism maybe considered illegal and by order of the law destroyed. The instantiation of the ‘Criminal Acts Policy’ is in particular important for this ‘Human Protection Policy’.
As stated in the beginning, the AI Leadership Framework is in an early stage and far from being complete. But in the interest of transparency also of our work, we wanted to share where we are at this stage and what we are working on. Suggestions may change into very different directions based on inputs and the work we are doing going forward.