Blog

ProjectBlog: Limitations regarding the development of A.I.

Blog: Limitations regarding the development of A.I.


“A.I.”, also known as Artificial Intelligence is a common buzzword these days and is often the domain of sci-fi speculation and imagination. With that being said, what is the true scope of A.I. and can it really be as powerful as popular depictions portray it to be? Or, is it more likely that A.I. will simply fulfill niche use cases and fall far from the kind of grandiose expectations that we’ve cultivated for it thus far? Perhaps the reality of A.I. will fall somewhere in between?

How A.I. currently works

A.I. currently depends on something called “Machine Learning”:

Machine learning is a method of data analysis that automates analytical model building. It is a branch of artificial intelligence based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention. (SAS)

Machine learning is basically where specially designed systems are exposed to large amounts of data from which they gradually learn to adjust and recognize certain patterns from it, this in turn constitutes the formation of their “intelligence”. The benefits to this particular approach are much clearer when compared to how A.I. was previously developed in the past via a programmatic approach. That is to say, in the past a programmer had to design an A.I. program with a specific response to every single kind of potential input. The limitations of this approach are pretty obvious given the scale and complexity of our world; could you imagine having to program an A.I. machine how to respond to every single situation? This doesn’t even take into account the near infinite amount of minute variations that every specific instance or situation entails which could easily derail or otherwise confuse a programmatically designed A.I. machine.

To give a very simple example of this, consider what we know as a “car”. As humans we have a basic, vague, yet powerful (for vagueness is where the human intellect derives its power) concept of what denotes a car. It is through possessing a basic mental template of what defines a car that we’re able to have a flexible mental model that can easily accommodate and adjust to any variation on what might constitute a car. However, imagine the difficulties in trying to translate this ability to a machine. How would it even be done? It’s apparent that the key to true artificial intelligence lies in the ability for a computer to have the ability to learn on its own. While this seems like an obvious enough answer (and it is), the technical means to actually execute such a concept were not available until relatively recently. It is only through the availability of new algorithms, improved hardware and greater availability of data (The fourth industrial revolution) that it’s become feasible to actually take the first few steps into both creating a computer capable of self-learning as well as providing the resources necessary for it to self-learn too (massive data-sets upon which the computer can practice pattern-recognition).

The weakness of machine learning

The problem with machine learning is that while numerous, different objects can be effectively modeled in the A.I. machine’s “mind” by exposing its data recognition mechanisms to large amounts of data; the A.I. machine doesn’t perform as well when exposed to entirely new situations. While humans have the ability to generalize and “export” understanding from a familiar situation to an unfamiliar situation in order to create an effective, relevant response; A.I. cannot currently replicate this ability. Therefore, in theory while A.I. can be trained to recognize lots of different objects, it only takes a single unfamiliar object or situation to “short-circuit” the A.I.’s ability to comprehend. Obviously given the complexity and randomness of the real world, this poses a significant problem to the development of A.I. In essence, this problem is basically a scaled up version of the original problem faced by using a programmatic approach towards developing A.I.; namely, how do you teach a computer to think abstractly and make generalizations? While it’s true that A.I. can now “intelligently” identify various objects, it still cannot reason. (AI Won’t Be Quite the Revolution You Expect)

Don’t worry, A.I. probably isn’t capable of this (yet)

How to solve this?

Complicated problems oftentimes require complicated solutions. Developing A.I. is no exception. As of now, the consensus seems to be that current ways of developing A.I. are gradually reaching diminishing returns and as a result this will require new ideas and new theories about how to tackle the “abstraction” problem. Therefore as it currently stands, we have reached a conceptual impasse about how to resolve this problem as opposed to a strictly technological one. Such a conceptual impasse will require creativity and ingenuity to learn how to resolve it as opposed to raw intelligence; ironically these requirements neatly mirror the very problems that they are seeking out to solve. Namely, how to get computers to think creatively as opposed to merely utilizing raw intelligence? I think that as the new, upcoming decade unfolds, innovative new ways of approaching this problem will inevitably present themselves in tandem with various other technological developments which will either inspire them, enable them or both.

Conclusion

A.I. as commonly depicted in popular media may still be a long ways off (or not), however even in the present day A.I. still plays significant roles, albeit in a scaled down, mundane manner. It will be very interesting indeed to see if (or when) A.I. makes the leap from Siri to C-3PO, however in the meantime we can be still be content with the many ways that common applications of A.I. continue to make our day to day lives more convenient. Specific examples of this which you may not have been aware were A.I. powered include Amazon.com which uses A.I. to more accurately predict and market particular items to shoppers, or perhaps how Pandora or Netflix is able to intelligently recommend media which viewers or listeners will enjoy (10 Powerful Examples Of Artificial Intelligence In Use Today). As you can see, A.I. is already making gradual inroads into our lives; it will be very interesting indeed to see the other ways in which A.I. could benefit us as well as the technology continues to progress into the future.

Sources

Adams, R.L. “10 Powerful Examples Of Artificial Intelligence In Use Today.” Forbes, Forbes Magazine, 6 Nov. 2017, www.forbes.com/sites/robertadams/2017/01/10/10-powerful-examples-of-artificial-intelligence-in-use-today/#467aa925420d.

Kelnar, David. “The Fourth Industrial Revolution: a Primer on Artificial Intelligence (AI).” Medium, MMC Writes, 2 Dec. 2016, medium.com/mmc-writes/the-fourth-industrial-revolution-a-primer-on-artificial-intelligence-ai-ff5e7fffcae1.

“Machine Learning: What It Is and Why It Matters.” SAS, www.sas.com/en_us/insights/analytics/machine-learning.html.

Adams, R.L. “10 Powerful Examples Of Artificial Intelligence In Use Today.” Forbes, Forbes Magazine, 6 Nov. 2017, www.forbes.com/sites/robertadams/2017/01/10/10-powerful-examples-of-artificial-intelligence-in-use-today/#467aa925420d.

Pontin, Jason. “AI Won’t Be Quite the Revolution You Expect.” Wired, Conde Nast, 2 Feb. 2018, www.wired.com/story/greedy-brittle-opaque-and-shallow-the-downsides-to-deep-learning/.

Source: Artificial Intelligence on Medium

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top
a

Display your work in a bold & confident manner. Sometimes it’s easy for your creativity to stand out from the crowd.

Social