Yiannis Maglaras // November 05 2018
What we actually know about AI?
Last week my colleague Anna Pastak has written a really interesting article about the possibilities and threats of AI from the standpoint of the everyday user of technology that I found quite insightful and refreshing. Especially refreshing in its cautious optimism and positiveness of what lies ahead that distances itself from the extremes of doom and gloom prophecies or blissful utopias that seem to monopolise the debate around AI.
The reality is that this sort of debate is not unique to Artificial Intelligence, but has been part and parcel of almost all technological advancements in the last two centuries. A very pertinent example is that of the railways. Before the railway opened, there were fears that it would be impossible to breathe while travelling at such speeds or that the passengers’ eyes would be damaged by having to adjust to the motion. Other Cassandras, including eminent scientists, expressed concern that cows disturbed by the noise would stop producing milk or that sheep would turn black from the smoke. We may laugh, now, but that doesn’t mean that railways haven’t affected our health in unpredicted ways. In order to efficiently run the railways, time became standardised because trains had to run to a set timetable across countries. Some scientists claim that the origins of stress in our modern society can be traced back to the introduction of exact time in the cities and towns of 19th century.
Mistaken extrapolations, limited imagination, and other common mistakes distract us from thinking more productively about AI powered future. Predicting the future is admittedly hard but if we are to have a go then let’s take some inspiration from how machine learning (field of Artificial Intelligence) goes about making predictions; by learning from known data and outputs. So what is that we know about AI?
1. Current state of AI allows for solving narrowly defined tasks
While it is far from reaching super intelligence, AI is getting better and better at accomplishing narrowly defined tasks. Businesses train models to deliver predictions (regression tasks), or extract insights from huge amounts of data (clustering), or help with classification tasks. We are not solving very complicated problems, except that all these basic problems are becoming building blocks of achieving something much bigger.
2. AI is already ubiquitous in our lives
Advances in machine learning algorithms have provided us with powerful toolset for solving problems that previously were deemed too hard, if not impossible, to deal with. This opened up great opportunities for all sort of businesses to improve products and processes and achieve significant competitive advantages. As a result, AI powered products and services empower many aspects of our daily lives; from our smart assistants like Alexa, recommendation engines on Netflix or Amazon to spam filters.
3. The democratisation of AI is putting powerful tools in the hands of everyone
Driven in particular by large cloud computing providers like Amazon, Google, and Microsoft, there are a growing number of tools to help beginners start to build their own machine learning models. These tools provide pre-built algorithms and intuitive interfaces that make it easy for someone with little experience to get started. Putting these tools in the hands of non-experts (it still requires considerable coding experience and a brain wired for data) could mean companies that don’t have the resources for the top data professionals can still reap the benefits of AI. The commercialisation of AI and the widening of AI adoption attracts even more funding to create a virtuous cycle that should accelerate advancements in the field of Artificial Intelligence.
4. The discussion about the ethics of Artificial Intelligence is gaining momentum
The year 2018 has been a turbulent year for Artificial Intelligence with negative stories coming thick and fast; Cambridge Analytica scandal, Google building AI systems for the US department of defence, fatalities of drivers and pedestrians from autonomous cars just to name a few. Although AI decision-making is often regarded as inherently objective, the data and processes that inform it can invisibly bake inequality into systems that are intended to be equitable. The idea that AI could function unaffected by bias reflects a misunderstanding of how the technology works. All machine intelligence is built upon training data that was, at some point, created by people. Moreover, competitive advantages delivered through AI can be used for the wrong purposes. All the above has fuelled the debate around transparency and ethics in AI and kickstarted intensive research in autonomous power with the ability to learn using assigned moral responsibilities.
5. AI has a long history of being the next big thing
The concept of AI was first proposed almost 70 years ago, but AI research has endured a bumpy journey and survived two major droughts of funding, known as “AI winters”. There is no better way to summarise the challenges of AI than the following sentence:
The hard problems are easy, and the easy problems are hard
Known as the Moravec’s paradox, it states that “It is comparatively easy to make computers exhibit adult level performance on intelligence tests, playing checkers or calculating pi to a billion digits, but difficult or impossible to give them the skills of a one-year-old when it comes to perception and mobility”. It’s hard to foresee what lies between where we currently are and Super-AI from a technical standpoint. The road towards Singularity may not be as straightforward as many tend to believe.
- The adoption of AI will be further accelerated
- AI algorithms that can learn multiple tasks and exhibit positive transfer of knowledge between tasks will be developed (i.e.: reinforcement learning) pushing the boundaries towards Artificial General Intelligence
- After reaping the low hanging fruits of AI research and development, new technological barriers may push back the realisation of Singularity (Super-AI) much further down the line than what’s currently expected.
- As scientists, institutions and governments acquire a better understanding of AI and its impact in multiple aspects of our lives, they will become better qualified to place the right checks and balances in place for reigning over the negative aspects of the technology.