Better Future through AI: Avoiding Pitfalls and Guiding AI Towards its Full Potential

by   Risto Miikkulainen, et al.

Artificial Intelligence (AI) technology is rapidly changing many areas of society. While there is tremendous potential in this transition, there are several pitfalls as well. Using the history of computing and the world-wide web as a guide, in this article we identify those pitfalls and actions that lead AI development to its full potential. If done right, AI will be instrumental in achieving the goals we set for economy, society, and the world in general.



page 1

page 2

page 3

page 4


Artificial Intelligence Approaches

Artificial Intelligence (AI) has received tremendous attention from acad...

Towards an unanimous international regulatory body for responsible use of Artificial Intelligence [UIRB-AI]

Artificial Intelligence (AI), is once again in the phase of drastic adva...

The Singularity May Never Be Near

There is both much optimism and pessimism around artificial intelligence...

The Short Anthropological Guide to the Study of Ethical AI

Over the next few years, society as a whole will need to address what co...

The potential of artificial intelligence for achieving healthy and sustainable societies

In this chapter we extend earlier work (Vinuesa et al., Nature Communica...

Aligned with Whom? Direct and social goals for AI systems

As artificial intelligence (AI) becomes more powerful and widespread, th...

ESR: Ethics and Society Review of Artificial Intelligence Research

Artificial intelligence (AI) research is routinely criticized for its re...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

After 60+ years, Artificial intelligence (AI) has moved from academic research discipline to a technology that affects people’s lives every day. We have digital assistants with which you can carry rudimentary conversations, systems that make medical diagnoses more accurately than humans, and cars that drive themselves in regular traffic, for instance.

At the same time, despite decades of development, AI is still in its infancy when it comes to commercial applications. There are few standards, little cooperation across companies and countries, and business users and consumers still rely on a small group of experts to be able contribute to AI solutions. There are significant issues that also need to be solved to ensure that as AI adoption grows, it creates positive effects on businesses and society. Like other technologies, AI is vulnerable to new security and privacy risks, and as a learning system, it is also subject to biases and potential abuses which can cause significant physical and financial damage. It is time to nurture AI towards maturity and responsibility.

Promoting and harnessing AI in the right way may seem like a daunting task at first. However, a crucial observation is that while AI is new and complex, it is not that different from other recent technologies that started as limited research projects and ended up becoming part of modern infrastructure, such as computing and the world-wide web (WWW). By analyzing how these earlier technologies developed, we can gain insight into where AI might be headed in the future. By identifying what went well and what went wrong, we may be able to identify the dangers of AI development and how to mitigate those dangers in the future.

To that end, this paper first reviews the history of computing and WWW, identifying four main stages of development in both: Standardization, Usability, Consumerization, and Foundationalization (Figure 1). The same four stages are then identified in the future of AI, and action recommendations are made based on lessons learned in computing and WWW. In the end, assuming that AI is allowed to developed in this guided manner, it is possible to reach a better future where AI plays a foundational role the same way computing and WWW do in our current society.

Figure 1: The four phases of development and impact of computing, web, and AI. Phase 1 represents the first step in commercialization. Companies rushed to monetize years of previous research and development with initial offerings. These offerings generally did not work with offerings from other companies. They could only be used by experts. Also, they were fragile when applied to business, requiring significant support and tuning. Companies adopted de facto and government standards to try to resolve these issues and grow the value of the initial offerings. Phase 2 represents a need to break past the early adopters and experts to get wider use in business. In this phase, companies focused heavily on usability of their products. In some cases, there was substantial copying of usability innovations, often without any interoperability between companies. User growth grew dramatically. Phase 3 represents the expansion of these technologies to consumers and individual innovators to create value. Access to app stores and self-publishing platforms dominated, and user-based content grew exponentially. Phase 4 represents a turning point where businesses built natively for the new technologies begin to dominate the market. In this phase, companies that did not adapt to completely embrace the new technologies failed, while newer or more adaptive businesses capture the bulk of the market value. In order to guide AI to realize its full potential, lessons learned from earlier technologies can be applied to AI.

2 The Case of Computing

Until early 1970s, computing technology was accessible to only a handful of individuals working in research institutions. In mid to late 1970s, the first personal computers were created [2, 3]. Initially they included many different architectures and operating systems, such as Altair 8800, Commodore PET, and Apple II, and TRS-80. Personal computing remained a relatively rare opportunity however, until IBM PCs were introduced in 1981. It had an open architecture, which made it possible for many manufacturers and software developers to build their own machines and systems [8]. This standardization resulted in PCs becoming commonplace and useful.

They were still difficult to use however, until graphical user interfaces were developed for the PCs. Although many had existed before, the Apple Macintosh in 1984 made such interfaces easy to use and available to ordinary users. The Windows system soon followed for PCs. Through these GUIs, it was no longer necessary to be a computer scientist to use computers. Such usability vastly expanded their applications.

The next phase was ushered in by smartphones, and iPhone in particular in 2007. While early smartphones such as Nokia’s Communicator and RIM Blackberry were miniature computers, iPhone instead provided an interface that hid the computing and focused on access to applications [9]. There were significant court battles of patents (Apple vs. Samsung, Apple vs. Qualcomm) which represented where innovations were creating substantial value. This phase resulted in consumerization of computing: anyone could now access computing at anytime, and most of the applications became computer-oriented instead of business oriented.

The fourth phase is happening now: Computing is becoming an infrastructure that is invisible to most people, like electricity or plumbing. It is accessed through numerous devices, including phones, cars, cashiers, homes, and much of it happens in the cloud instead of locally. People do not have to care where and how it happens—they simply interact with its results, the same way we interact with a light switch or a faucet. Computing is a foundational infrastructure upon which we build our everyday activities. On the other hand, it is dominated by only a few players (AWS/MS/Google) that thereby have a significant control of those activities.

3 The Case of World-wide Web

The development of the World-Wide Web followed a remarkably similar trajectory. In the early 1990s, it had became possible to distribute information over the internet, including services such as ftp sites, Usenet news groups and gopher servers. However, they were accessible only to the initiated, and were of limited use. The first stage, standardization, occurred with the invention of the HTML protocol [1]. It made it possible to build content in a common format, and access it worldwide using HTML readers, i.e. browsers.

The content was still mostly text, and therefore of limited use. The invention of style sheets [6] made the WWW vastly more flexible and usable. Separating presentation and content, it was possible to present information in a visual manner that made the content accessible to the wider population. It became possible to develop web interfaces for businesses, and information in general became accessible through the web, in addition to traditional media.

The third step became known as the Web 2.0: instead of being passive consumers of information, anyone could now contribute to the WWW [10]. People started to put much of their lives in the WWW through social media platforms such as Facebook. It become possible to create content such as blogs and videos, and find an audience through YouTube and other similar sites.

The fourth phase is happening now, and its implications are huge. The WWW has become a standard infrastructure for commerce and creativity. It is possible for small businesses to reach consumers across the globe. Brick and mortar stores have become secondary in many areas of retail such as books, apparel, and groceries. Travel, entertainment, and everyday life in general is organized through the Web. As a result, WWW is the foundation of most human activity in the modern world, making our lives more rich and efficient than ever before. On the other hand, fake news now propagate easily, social computing and gaming can be considered an addiction, and outages of social media service causes an outrage, demonstrating that our relationship with the WWW is complex and still evolving.

4 Lessons for the Future of AI

In the light of the above two examples, let us examine how AI is likely to develop in the future. We are at the early days of AI similar to computing in the 1970s or WWW in early 1990s: Many examples of AI successes exist but they are disjoint and opaque, and accessible and understandable to experts only. The four phases that are likely to follow are outlined below.

4.1 Phase 1: Standardization

Like the open architecture of PCs and the HTML of the WWW, we need open standards for AI systems. Such standards will make it possible to build interoperable AI systems, i.e. to build on successes of others. For instance, it will be possible to connect a language generation system to a vision system, and then a translation system to generate in another language, and a speech generation system to output the result. Through standards, it should be possible to transport the functionality from one task to another, e.g. to learn to recognize a different category of objects. It should be possible to swap modules of a system in and out, such as replacing one language with another. Such standardization will leverage the successes of current AI and make many more applications possible.

An important aspect of standardization is also trustability, which can be seen as a generalization of interpretability and explainability. Ultimately, more important than having AI explain explicitly what it is doing is that we can trust its decisions. Much of very useful AI is based on statistical inference which can be opaque. We should not rule out such AI because there are no simple linguistic explanations. Instead, we should establish standards of trust, i.e. ways to certify that the behavior of AI is fair and unbiased, that it knows its limits, and is safe, whether its behavior can be explained in simple terms or not. Such standards of trust make it possible to apply AI safely to many more situations.

4.2 Phase 2: Usability

Like GUIs and stylesheets made computing and WWW accessible to non-experts, AI needs interfaces that make it possible for everyone to use them. An important lesson comes from browser wars of the late 1990s [4]. Initially there had been rapid development and innovation among web browsers. However in late 1990s, Microsoft gained a dominant position by bundling its Explorer to Windows, and in essence cutting off competition. As a result, innovation stopped for several years, until the antitrust case and mobile computing got it going again.

The lesson is to ensure that in the development of AI for the larger market, open competition and innovation must be ensured. It should not be possible for one player to force adoption of their AI technology simply because they dominate some other part of IT. Note that standards help in this aspect as well, making it possible for new innovations to interoperate with existing ones, instead of making AIs incompatible. The result will be open innovation in creating AIs that will be useful for the general population.

4.3 Phase 3: Consumerization

Like iPhone made computing available to everyone everywhere, and Web 2.0 made it possible for anyone to contribute to the WWW, in this phase it will become possible for anyone to build AI applications to their needs and for general consumption. This means mass production of AI-based systems by the general public: people can routinely produce, configure, teach, and such systems for different purposes and domains. They may include intelligent assistants that manage an individual’s everyday activities, finances, and health, but also AI systems that design interiors, gardens, and clothing, maintain buildings, appliances and vehicles, and interact with other people and their AIs.

A lesson learned from the recent problems with privacy and fake content [11, 5] suggest that there is a danger that this process will run amok. However, there is also great potential in encouraging creativity and enriching people’s lives. If Facebook and YouTube had been moderated, editorialized, and regulated from the beginning like traditional media, it is unlikely that such creativity would have flourished. We need to be able to watch and learn from such unbridled AI development, and avoid regulating it until it becomes absolutely necessary. Only then it will be possible to harness the potential of democratized AI innovation.

4.4 Phase 4: Foundationalization

The way computing has become invisible and WWW has become a primary means of interaction, so will AI become ingrained into the society. It means that AI will be routinely running business operations, optimizing government policies, transportation, agriculture, and healthcare. This does not mean that human decision making is replaced by machines—it means that human decision making is empowered by machines.

More specifically, AI in the future is not limited to prediction, but it can also prescribe what decisions need to be made to achieve given objectives [7]. But only humans can define those objectives—we cannot delegate them to AI agents. At the macro level, we will need to decide what kind of society we want to live in, and derive the objectives accordingly. For instance, we may decide to maximize productivity and growth, but at the same time minimize cost and environmental impact, and promote equal access and diversity. AI can then be directed to discover ways in which those objectives can be achieved.

In past and current societies decision making is often obscured by special interests, historical inertia, and personal agendas, and consequently it has been difficult to prevent conflicts and promote opportunity despite best efforts. In contrast, AI in this fourth phase will provide the tools to bypass such factors and build a society we want to build. Thus, for the first time in history, we will be in control of our own fate.

5 Conclusion

AI is the technology that makes a better future possible. A common misconception is that AI is something uncontrollable that leads to disasters or will eventually take over. Given the process above, it is instead something that will be developed by humans in service of humans. AI will eventually become powerful enough to run much of the society’s infrastructure, but it will only get there through the phases outline above. Each step of the way leads to more powerful AI that serves the humanity better. Our job is to guide its development to make this process productive and safe.