How has Artificial Intelligence evolved over the years?
Over the years Artificial Intelligence (AI) has evolved considerably. We have seen progress made in several fields: image recognition, natural language processing and autonomous vehicles to name a few. While the industry is still a long way from strong AI (a system that could successfully perform any intellectual task that a human being can), many of the developments in narrow AI have resulted in systems that can collaborate with individuals within organizations. Robots in manufacturing settings are driven by sophisticated expert systems that lead to better and less expensive production. In the medical field, AI systems can complete the first pass in reading X-Rays, freeing-up radiologists to spend their time on more complex cases. In the military, smart weapons are making soldiers more effective.
These advances in AI are being powered by three main drivers:
With the gradual slowing down of Moore’s law (exponential growth of general CPU performance), the high-performance-computing community are searching for alternatives.
It is now widely accepted that the GPU (graphical processing unit) with its 1000s of elemental cores is better suited for the multi-layer neural-networks being used by deeplearning algorithms. Several other computing alternatives being touted for AI are: the FPGA (field programmable gate array), ASIC (application specific ICs) like the TPU (Tensor Processing Unit) from Google and special hardware projects like the AP (Automata Processor) from Micron. Ultimately end-to-end AI systems will make use of a combination of these hardware engines.
Supervised learning systems based on neural networks generally improve their error rates with either more labelled data (bigger training sets) and/or deeper and wider layers of the network. Organizations have begun to understand the value of analysing their collected data about their operation, customers and products. Adding AI to the mix, often accelerates the benefits of this data especially if it is labelled and can be used for training, validation or testing.
This has resulted in a focus on generating bigger data sets with some applications built only for their ability to collect more data.
Neural networks with feed forward and back propagation continue to be the basic building block of AI algorithms. For example, convolution neural networks (often referred to as convnet) are widely used for image processing. AlphaGO’s landmark win against one of the best GO players used a combination of reinforcement learning and deep neural networks.
Progressive neural networks have been developed as solutions that drive advances in gameplaying AI. Another recent approach is called generative adversarial networks(GANs) which
combines two neural networks that feed each other with realistic synthetic data resulting in
a stronger more adaptive algorithm over time.
Progress in these three areas is being regularly announced. This is helping drive the evolution of AI as highlighted by the new products and services being launched by the various AI projects in the technology community as well as both the incumbent and start-up vendors in numerous industries.
What are the top challenges future leaders of Artificial Intelligence are facing today?
In my opinion, future leaders of Artificial Intelligence(AI) face two main challenges. The first is the hype around AI especially after a big event in which a human expert is shown to be inferior to an AI system. The second is the back lash from various groups who believe that AI will have a negative impact on humanity. Below, I explore each of these challenges separately.
There is a long list of examples in which AI is predicted to achieve the Turing Test which dates to the 1950s when the term was first coined. For example, in 1959, it was reported that a computer could be programmed so that it will learn “to play a better game of checkers than can be played by the person who wrote the program”. This led to conclusions that soon we would have programs that learn from their experience as effectively as humans do. In that same time-period, General Motors introduced the first industrial robot on the assembly line. It was predicted that by 1985, “machines will be capable of doing any work a man can do”. While this hype made for good headlines, it set back a lot of the research funding and interest when advances were much slower than anticipated. In fact, it is widely accepted that the 1970s was the first winter for AI.
After some progress on expert systems and the creation of humanoid robots, the resulting hype was responsible for a second winter for AI in the late 1980s. Even though some important progress was made with the development of back propagation neural networks and language processing, there was call for the singularity for superhuman intelligence to be reached by 2023. When Deep Blue beat Kasparov (the reigning world chess champion) in 1997, the hype seemed justified. Twenty years later AI is making progress (Watson winning at Jeopardy in 2011) but most researchers acknowledge that it is still very far away (Watson was faster on the buzzer) from achieving the multi-tasking performance of humans. Along with the hype, AI brings predictions of negative impacts on society. While it’s true that the three previous phases of the industrial revolution (steam, electricity and computer) have disrupted the job market, the negative repercussions have decreased significantly with each phase. The major reason is that robust societal institutions have been created to reduce the impact that paradigm shifts have on society. Already we are witnessing some progressive governments (e.g. Netherlands and Finland) experimenting with universal basic income programs to ensure that income being lost to AI automation does not result in social unrest. The challenge that future leaders will face is to change their business mind-set from short term profits to long-term sustainable revenue streams. Additionally, there are many scare mongering books and articles being written about the
destruction of humanity, when super-intelligent robots take over the world. While there is a
remote possibility that this may happen, the human race has developed sophisticated world
bodies that are staffed by level-headed logical thinkers. We have been able to control nuclear weapons, we are making progress on global warming, so in my opinion we should be able to do the same with artificial intelligence. The singularity of super intelligent machines is not yet a forgone conclusion, but future AI leaders will need to provide answers to some of these issues as they continue making progress in their field.
If there is advice for key leaders to prepare for the new world of Artificial Intelligence,
what would it be?
Artificial Intelligence (AI) is still in its initial phase of implementation in society and industries. Its impact is predicted to be as powerful as electricity was 100 years ago. There are three areas that key leaders are going to need to address to either minimize the disruption or maximize the benefit of this paradigm shift:
Although we have had several false starts in AI, over the last ten years significant progress has been made with the technology. Every week, a new use-case is announced that has the potential to make a significant impact on a specific industry. Leaders must remain up to date on the technology and constantly evaluate how important AI is going to be to their business moving forward. The outcome of their decision should lead them to assign an owner with a budget that reflects the priority AI is going to have in their future. The AI owner should report to a C-level executive if not directly to the board.
As part of their decision, leaders need to reflect whether they are going to be leaders or followers for this technology. A good practice is to start with a constrained well-defined automation project. The business plan may call for them to automate proven areas of their back-end operation but not use it to engage with customers at the front-end. This is a good way to get familiar with the technology but ultimately leaders must explore how AI can provide competitive advantage when it is combined with their intellectual property and organizational structure. This may often require a fail-fast innovation approach.
The temptation to use AI to reduce or replace the workforce is going to be quite compelling.
While in theory this may result in short term profits, in my opinion it will not be a sustainable business model. Other firms will copy the model which will squeeze profits and without the social capital of a strong workforce, the firm will not be able to respond to the competition. Having a mind-set to use AI to reduce the mundane tasks of an employee to allow them more time to focus on addressing challenges and creating new revenue streams is much more of a sensible approach.
Overall business leaders are going to be challenged by AI, but like any major shift in an industry, it is those leaders that resist taking a reactive short term response that will be successful at navigating these challenges.
BACK TO NEWSROOM
Asia Cloud Computing Association
MEET BERNIE AT THE CONFERENCE
SPONSORING / EXHIBITION
Romain DECHAUDAT • Sales Director
DID: +33 (0) 1 58 12 06 94
Jared CHIN • Conference Manager Singapore
DID: +65 6265 3532
Mobile: +65 8116 4877
AI asia 2017 by
Corp Agency is an international corporate event management firm specializing in the organization of first- class conferences, trade shows and business meetings in the field of innovation and technology. We ensure top quality content and speakers for attendees, innovative business meeting solutions and exceptional networking opportunities. Corp Agency has a wide presence in Europe and America and is currently expanding in Asia.