What is Artificial Intelligence (AI)?
There are many definitions of Artificial Intelligence (AI) but none of them effectively captures what AI is capable of. As such, I will not be defining AI, rather, I will be reporting the goal of AI as encapsulated at the 1956 Dartmouth Summer Project on Artificial Intelligence, where the science and technology of AI was properly born.
It goes thus:
‘To proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can be so precisely described that a machine can be made to simulate it’[i]
Although it has hardly been restated, this goal has the been the underlying drive behind every manifestation of AI since then. If it takes human intelligence and learning to get a thing done, then it can be replicated by a machine, once we break down the components of the particular intelligence and learning approach at play. It took a high level of numerical intelligence and consistent practice for Gary Kasparov to become the world chess champion.[ii] By replicating the rules of the game, approximately 200 million possible chess positions and reducing them to algorithms, IBM’s Deep Blue was developed and trained until it defeated the world chess champion.[iii]
Why is a discussion of AI in warfare important?
This essay is premised on several real and assumed fears about the application of AI. In the news, we hear about China’s credit scoring system that ranks citizens based on metrics not publicly available. Those who fall short can be legally denied the most basic amenities, such as a train ride, access to the best schools, or being publicly named as a bad citizen.[iv] In China, AI has made the police state a reality, not just something confined to the pages of George Orwell’s ‘1984’.
In the USA, we hear of risk assessment systems like COMPAS that have delivered recommendations skewed against persons of colour, as a result of data gaps.[v] Clearly, the use and abuse of AI is no respecter of economic and political ideology. All nations who can afford to, have shown that they have the tendency to use AI wrongly.
When you combine these factual occurrences with fictional notes form popular movies and books about robots supplanting and almost destroying the human race, it’s easy to understand why the concept of ‘Artificial Intelligence (AI) in Warfare’ may be greeted with trepidation. War is already a costly business, but at least we can hold humans responsible and make them account for their misdeeds. But how do you negotiate a truce with a weaponized drone gone haywire or an artificially intelligent missile guiding system?
What are the real factors that necessitate these fears?
The examples earlier given may seem farfetched unless you consider the fact that security agencies world over have been engaged in using advanced technologies. In the years after 1956, the US Government, through the Defence Department and associated agencies invested heavily in AI research & development.[vi] This trend continued through till the 70’s when public pressure mounted against this, as politicians and the people felt that the funds should be spent on more beneficial activities. This led to the period known as the AI Winter, when funding for AI research & development became scarce and far between.
Although it only kicked off in the late 90’s due to private interests that have built and sustained the AI industry to its current levels, renewed government interest in AI has once again gone in the direction of security purposes, rather than solving numerous social challenges like climate change, food security, education, healthcare, etc. Recently, it took the public opposition of ‘Googlers’ (the term of reference for people who work at Google) to stop the company from renewing its contract with the Defence Department to work on the Maven program, a system that uses AI to interpret video images for improving the targeting of drone strikes.[vii]
We might praise the company for choosing to heed its workforce, thus displaying great company culture, but the fate of the world cannot be hinged on isolated good acts. Another company could have chosen to ignore such calls. Google could afford to pass over the opportunity because it is big enough, but a smaller company may not. Even the government in question will not stop because one organization turned it down.
As long as governments the world over continue to see AI within the narrow prism of a tool to be used to enhance their security capabilities, the use of AI in warfare is not just a potential threat, but an actual long-term inevitable occurrence.
Are there benefits to be derived from AI in warfare?
It is worth considering the benefits that AI in warfare could bring. By doing so, I am not endorsing war, but since it is a potential outcome of human conflict, the possibility of machines taking the place of man in warfare is worth considering.
Since World War 1, when advances in technology began to play a central role in who had the upper hand, millions of lives have been lost on the side combatants and innocents alike. Developing artificially intelligent robots that could take the place of human military personnel would significantly reduce or eliminate the unnecessary human casualties that war produces. Wars could be finished in shorter timespans, rather than dragging on for years. Human troops require feeding, rest, consistent morale boosting and several other factors which may not guarantee victory. But a robot army, navy, or air force could fight 24/7, or as long as their power sources last, irrespective of external factors.
Also, machines would most likely show respect to the civil rights of non-combatants as opposed to human soldiers. No machine intentionally would rape an innocent woman, line civilians up for a firing squad, wipe out entire villages, or destroy national monuments. All these atrocities have one thing in common; humans committed them. So, is there a case for Artificial Intelligence (AI) in Warfare? Yes, there is.
Over the years, the tide of convention drafting has veered towards regulating the activities of men in the course of war. There are several of them, but I will make mention of the underlying principles that will be advanced by the foray of AI in warfare. They are as follows:
- That war should be limited to achieving the political goals that started the war and should not include unnecessary destruction
- Wars should be brought to an end as quickly as possible
- People and property that do not contribute to the war effort should be protected against unnecessary destruction and hardship.
If these are the minimum objectives that the laws of war seek to achieve, then AI can be effectively used to ensure that they are completely fulfilled.
Robot soldiers or weapons could be programmed only to harm enemy personnel, while leaving the civilian population untouched. If wars were only focused on the armed forces of opposing nations and not on civilians and properties, they would be prosecuted in shorter timeframes. The needless destruction, hardship, and economic turmoil that usually follows war when people and their livelihood is destroyed would be averted.
How can understanding the types of AI help us appreciate its relevance in warfare?
In understanding the place of AI in war, a deeper understanding of AI needs to be gained. Broadly speaking, AI can be divided into ‘Weak AI’ or ‘Narrow AI’, and ‘Strong AI’ or ‘General AI’. Most of the manifestations of AI we see today are ‘Weak AI’. It is so-called because they replicate humans in a specific set of tasks like the way Siri, Alexa, and Cortana, all virtual assistants, can answer your questions, provide you directions, but cannot operate a car. A particular data analytics system may be able to interpret data to give recommendations on whether a person should be given a loan, but that same system would be unable to give an assessment on whether the person should be granted bail or not.
‘Strong’ or ‘General’ AI only exists now as a theoretical concept. It is based on the theory that autonomous agents can be developed which can perform a wide variety of distinct activities just like the average human. Imagine all the AI products and services you know rolled in one and you have a rough idea of what Strong AI is.
It is important to bring these distinctions to light in this essay to show how AI can be effectively used now. Currently, AI can only be developed in narrow terms. It is thus possible to develop AI tools for war that perform very specific functions and will be incapable of violating the laws of war because of their algorithmic makeup.
How will AI affect the laws of war?
Doing so will however not come easy. The rules in their current form accommodate the application of autonomous weapons, within the context of Article 36, Protocol 1 of the Geneva Conventions 0f 1949, which provides that new weapons, means, or method of warfare, must be tested by a High Contracting Party to determine whether its employment would in some or all circumstances be prohibited by the Protocol or any other international applicable law available. Specific updates may still have to be made to effectively regulate the deployment of AI in warfare.
AI may replace humans wholly or in some specific assaults of war. Because the laws that regulate these aspects are tailored towards humans and impose liabilities simply extending them to AI tools may pose challenges. For instance, it would require that we assume that AI has some form of consciousness and legal personality in order to hold them liable for the consequences of their actions, in that case.
Another challenge would be defining the limits of AI in warfare. With humans, a simple propaganda speech or a series of meetings would be adequate in getting the soldier informed about the objectives in a war. But with AI, this process cannot be applied; it is most likely that the principal objectives would either be programmed into it or transmitted by a server. In some other cases, it is possible for them to perform actions that even their makers did not envisage, as they get more autonomous. Once put in the real-world environment, the advanced AI system could learn from its environment, thus leading it to act in a way that its designers did not deem feasible. We saw a glimpse of this in the chatbot that Microsoft once launched on Twitter. In less than a day of being launched, it had learnt from the Twitter environment and was spewing racist, sexist, and hateful comments on Twitter, none of which were anticipated by Microsoft, so they deleted it.[viii] These kinds of AI systems are self-imbued with the ability to continue modifying themselves, improving themselves from constant interactions, such that their behaviour is based on the new data they have been exposed to, not the original instructions of the developer. AI tools will largely perform tasks they have programmed to do. If in the performance of these tasks, they commit violations of the laws of war, it becomes necessary to consider where liability will lie.
What is the way forward?
To solve the challenge that AI poses if introduced to warfare, there are several approaches we can take. First, we could decide to expressly ban the application of AI. This precautionary approach might prove effective, but like all blanket bans, it would also impede the development of some forms of AI that could otherwise be beneficial to humanity.
Secondly, we could decide to not stand in its way, and allow all nations of the world with the interest and capacity to develop AI weapons to do so. Towing this permissive route would imply that we accept the risks and potential fallouts, but we would be faced with the challenge of how to apply our international laws of war, especially when their use leads to injury.
We could also go all out and impose a strict liability regime on the developers and users of these AI systems, such that they would be held responsible for all fallouts of their machines, irrespective of circumstances. This approach is by all means, the simplest of all, and would not require much out of the box thinking to bring to reality. We would simply be extending already established legal principles to a new subject matter. This approach, just like the first, might also stifle he development of AI that could greatly enhance the human condition.
Another way we could do is to accept the challenge that AI poses, not just as a tool of war, but as a tool in itself for whatever purposes it is applied to; a tool with the potential of outworking and outsmarting its masters. By accepting the challenge, it poses, we would commit to rethink our entire legal and ethical framework, in other to discover new definitions and principles that would be applicable, without unnecessary pain or gain in the coming Singularity Age, when machine intelligence is expected to match and possibly outperform human intelligence.[ix]
Whatever approach we prefer, we must ensure that we keep three things in mind:
One, that AI is first a tool for enhancing our
understanding and improving our world. Two, looking at AI from the narrow prism
of a tool for advanced warfare is as unfair as failing to prepare for its use
in warfare, which is inevitable. Three, Artificial Intelligence like a mirror,
reflecting all that is good, bad, and ugly about us, and throwing all this at
us. So before you point an accusing finger at the technology, remember that
there are four fingers, pointed back at you.
[i] Program of Events, AI@50 – Dartmouth College Artificial Intelligence Conference: The Next Fifty Years, July 13 – 15, 2006, pg. 1
[iv] Alexandra Ma China has started ranking citizens with a creepy ‘social credit’ system – here’s what you can do wrong, and the embarrassing, demeaning ways they can punish you (Business Insider – October 29th, 2018) www.businessinsider.com/china-social-credit-system-punishments-and-rewards-explained-2018-4
[v] Carole Piovesan and Vivian Ntiri, ‘Adjudication by algorithm: The risks and benefits of artificial intelligence in judicial decision-making’, The Advocates Journal, Spring 2018, pgs. 42 – 45
[vi] ‘DARPA Announces $2 Billion Campaign to Develop Next Wave of AI Technologies’ (Defence Advanced Research Projects Agency – DARPA 7th of September 2018) https://www.darpa.mil/news-events/2018-09-07
[vii] Daisuke Wakabayashi and Scott Shane, Google Will Not Renew Pentagon Contract That Upset Employees, (The New York Times, June 1st, 2018) www.nytimes.com/2018/06/01/technology/google-pentagon-project-maven.amp.html
[viii] James Vincent, Twitter taught Microsoft’s AI chatbot to be a racist asshole in less than a day, (The Verge, March 24th, 2016) www.theverge.com/platform/amp/2016/3/24/11297050/tay-microsoft-chatbot-racist
[ix] Roey Tzezana, Singularity: Explain It to Me Like I’m 5-Years-Old, Futurism, www.futurism.com/singularity-explain-it-to-me-like-im-5-years-old/amp