Parnas, D. L. (2017). The Real Risks of Artificial Intelligence: Incidents from the early days of AI research are instructive in the current AI environment. Communications Of The ACM SUNNY ANSARI

From Digital Culture & Society

The ongoing increase in speed, memory limit, and correspondences capacity enables the present PCs to do things that were unbelievable six decades prior. At that point, PCs were fundamentally utilized for numerical estimations; today, they process content, pictures, what's more, solid accounts. At that point, it was an achievement to compose a program that played chess badly however accurately. The present PCs have the ability to rival the best human players. The fantastic limit of the present computing systems enables a few purveyors to depict them as having "artificial intelligence" (AI). They assert that AI is utilized as a part of clothes washing machines, the "personal assistant" in our cell phones, self-driving autos, and the monster PCs that beat human champions at complex amusements.

The article written by David Lorges Parnas addresses the risks and dangers that are associated with the technology that we define as “artificial intelligence”. As many scholars have been researching upon the benefits and opportunities that AI proposes, this writer takes a unique approach to the topic, as he discusses the dangers that appear when artificial intelligence does not work in favour. A primary concern is the replacement of humans in vast industries that can be vanished by the power of artificial intelligence. The article addresses a quote from a Microsoft researcher explaining “As artificial intelligence becomes more powerful, people need to make sure it’s not used by authoritarian regimes to centralize power and target certain populations.” (Parnas, 2017). Automation has fundamentally changed our general public, and will proceed to do as such, however the worries about "artificial intelligence" are unique. Utilization of AI strategies can prompt gadgets and systems that are dishonest and infrequently perilous.

The article addresses his time studying at university where he compares the uses of artificial intelligence. The lectures that were compiled through an AI software, were not seen to be as measurable and educational as the traditional lecturers with university professors. As teachers will properly teach their students to analyze problems thoroughly, artificial intelligence relies almost entirely on intuition. The article then states three distinguished types of AI research:

1. Building programs that mirror human conduct keeping in mind the end goal to understand human reasoning.

2. Building programs that play games well.

3. Demonstrating that practical computerized products can utilize the methods that humans utilize.

To further explain these types of artificial intelligence research, the article writes a column upon the early days of AI, that makes these dangers relevant in todays society. A well known point in early AI research and courses was the character recognition issue. The objective was to compose programs that could distinguish hand-drawn or printed characters. This task, which the greater part of us perform easily, is troublesome for computers. The optical character recognition software that companies use to perceive characters on an examined printed page every now and again blunders. The way that character recognition is simple for humans yet at the same time troublesome for computers is utilized to endeavour to shield programs from signing on to Internet sites. For instance, the site may show a CAPTCHA that is difficult to recognize a "8 and B”. Early AI specialists showed us to configuration character recognition programs by meeting human perusers. For instance, perusers may be asked how they recognized a "8" from a "B." Consistently, the guidelines they proposed failed when executed and tried. Individuals could carry out the activity yet couldn't explain how. Current software for character recognition depends on confining the textual styles that will be utilized and breaking down the properties of the characters in those textual styles. Most humans can read a content, in another text style without concentrate its characteristics, yet machines regularly can't. The best answer for this issue is to stay away from it. For writings made on a PC, both a human readable picture and a machine-readable string are available. Character recognition isn't required.

Another way to deal with AI depends on displaying the brain. Brains are a system of units called neurons. A few researchers attempt to deliver AI by impersonating the structure of a brain. They make models of neurons and utilize them to reenact neural networks. Counterfeit neural networks can perform basic tasks yet can't do anything that is impossible by ordinary computers. For the most part, ordinary programs are more productive. A few trials have demonstrated that ordinary numerical calculations beat neural networks. There is instinctive interest to building a simulated brain in view of a model of a biological brain, however no motivation to trust this is a functional method to tackle issues.

At whatever point designers discuss AI, make inquiries. Despite the fact that "AI" has no by and large acknowledged definition, it might mean something particular to them. The expression "AI" clouds the genuine system in any case, while it regularly covers up messy and deceitful techniques, it may disguise a sound system. An AI may utilize sound rationale with exact data, or it could be applying “statistical inference” of far fetched provenance. It may be a very much organized calculation that can be appeared to work effectively, or it could be an arrangement of heuristics with obscure confinements. We can't believe a gadget unless we know how it functions. AI strategies are slightest hazardous when it is worthy to get an off base outcome or on the other hand no outcome by any means. On the off chance that you are readied to acknowledge "I don't understand" or a unimportant reply from an "personal assistant," AI is innocuous. In the event that the reaction is imperative, be reluctant about utilizing AI. Some AI programs quite often work and are risky on the grounds that we figure out how to rely upon them. A failure may go undetected; regardless of whether failures are identified, clients may not be proceed without the device.

The article ends of with a statement that summarized this article strongly, as it leaves the reader to wonder more about artificial intelligence. “Instead of asking “Can a computer win Turing’s imitation game?” we should be studying more specific questions such as “Can a computer system safely control the speed of a car when following another car?” There are many interesting, useful, and scientific questions about computer capabilities. “Can machines think?” and “Is this program intelligent?” are not among them. Verifiable algorithms are preferable to heuristics. Devices that use heuristics to create the illusion of intelligence present a risk we should not accept.

References 1. Parnas, D. L. (2017). The Real Risks of Artificial Intelligence: Incidents from the early days of AI research are instructive in the current AI environment. Communications Of The ACM, 60(10), 27-31. doi:10.1145/3132724