Given the proliferation of various AI-related technologies into our lives, -ranging from smart robots to machine learning- our past disbelief in software being able to automate things that were definitely within the area of expertise of individuals was shaken completely. Although we may not have reached the peak of AI yet, there are some important factors that need to be taken into account when considering the impact of these technologies during its peak.
- The entry barrier for a developer: If developers are willing and motivated to create something useful, rapid dispersion of the technology happens so that improvements occur given the contributions of the developers. To give an example, there is a high barrier to entry for operating systems due to the complexity of the software. Therefore, not many developers try to revise an operating system for some specialized purpose.
- The centralized nature of development on the core platform: Companies may act as gatekeepers for contributions to the main platform. To give an example, within the social ecosystem, companies like Facebook have centralized the contributors.
These two factors are not necessary for technology to make an impact as other factors such as market consolidation or the level of innovation will also determine the direction of a particular technology. To give an example, despite the few number of players in the field of cloud computing, it has had a huge impact on the public.
O’Reilly identified the technology innovation potential based on a ranking of major technologies. Let’s go through what each quadrant represents.
Limited innovation: Innovation is limited due to the fact that, although developers can easily develop, they have to work within the boundaries of the platform owned by a single authority; therefore they may not be able to shape the direction. The app store can be considered an example within this quadrant.
Proprietary innovation: In this case of the conventional proprietary model, developers cannot contribute to the platform directly as only the platform owner is eligible to contribute to the platform.
Incompatible innovation: Technologies based on open source, or open based on complicated software or hardware, are within this quadrant, as the barrier entry for developers would still be very small. As a result of this, developers can freely embed the standards of their own choice, which may result in incompatible solutions.
Fully distributed innovation: Being one of the ideal models of innovation, developers within this quadrant can easily develop their own platforms. Yet, the lack of a central coordination usually leads to a fragmented ecosystem given the incompatibility among the frameworks.
Looking at one of today’s most popular trends -namely AI-, it is obvious that the entry barrier is low, given the freely available tools. Yet, what matters is a crucial data set, as well as the technical knowledge to develop a useful model.
One of the most distinguishing features of AI is the open publication of research studies with a limited amount of code in comparison to other fields of technology where the norm has been to maintain a closed model of innovation that entails a big amount of code. Behind the logic underpinning this difference of AI lies the idea of delivering the key point to the community as fast as possible so that they can, later on, improve upon it.
Another distinguishing feature of AI is the lack of existence of a main governing institute which would be eligible to approve certain standards rather than allowing these standards to be enforced by certain big powers within the industry.
A confluence of several circumstances has led to the booming of AI, such as:
- A fully dispersed innovation environment;
- The development of big data infrastructures by companies;
- The willingness of companies to see a value realization in big data investments by seeing big data as a means to an end;
- The readiness of optimized computing for AI given the capabilities of GPUs for matrix multiplication which is much faster than CPUs;
- The low-level barrier to entry to develop AI solutions, although there is the access requirement to significant data.
In order to make AI systems flourish, the following external factors should be taken into account to overcome potential impediments:
- A deep focus on deep learning: Given the attention that deep learning has attracted, we may have unnecessarily focused on it which may impede us from looking for other methods that may be helpful in achieving our goals for AI.
- The lack of quality in big data: Although most companies may be good at organizing their data, they still are in lack of quality data even though they think they may have better data than in reality.
- The human factor: As is the case in any other area of life, the biggest obstacle to achieving our goals in the field of AI may be ourselves given our reluctance to fully embrace the AI technologies which may save millions of lives. A recent example is the use of self-driving vehicles given the societal and political pressures.
- Surviving the Trough of Disillusionment: Despite the use of well-known Gartner curve of technology adoption for every technology trend, a different slope should exist to its curve for each emerging technology including AI. What if the famous “trough of disillusionment” for AI turns out to be much deeper than initially expected?
- Centralized development on emerging AI platforms: Given the popularity of Tensorflow, chances are high that it could be accepted as a default framework by AI engineers. In this case, the risk would lean towards a more centralized version of innovation.
According to Amara’s Law, while the effect of a particular technology is overrated in the short run, it is underestimated in the longer term.
Time will show us whether this law would come true for AI as well and whether there is a tipping point where the rate of innovation proceeds faster while additional breakthroughs occur.