Can computers really learn?

Can computers really learn?

It was the father of computer science himself, Alan Turing, who once said that, ‘if a machine is expected to be infallible, it cannot also be intelligent.’ Decades ahead of his time, Turing had already anticipated the difficulties that would come with developing artificial intelligence. His theory would later become the bases of modern machine learning algorithms, a concept which suggests that computers are capable of learning from data. But his predictions were correct because there are limitations with such advanced concepts and we’ve only just started to scratch the surface of possibilities.

How far have we come so far?

Machine learning is applied to various industries such as retail, publishing, banking, social media, healthcare and the financial sector. As a result, colossal platforms like Google and Facebook are able to accordingly advertise to consumers based on their search history or previous purchases. This reduces search time and increases use of resources. Of course, there are multiple algorithms for different platforms and uses because as the No-Free-Lunch theorem suggests: there is no one algorithm that works for every machine learning problem. So, the more available and varied the data is, the better it understands human behaviour.

But here lies the problem: the algorithms rely on data that has been created by humans where biases are organic.  Examples of this in AI is Tay, the chatbot created by Microsoft in March 2016, which had to be taken down 16 hours after it’s launch due to offensive comments made through its Twitter account. Initially, the bot was developed to become ‘smarter’ as more users interacted with it. But the opposite happened as Tay quickly learned how to mimic the chauvinistic nature of human twitter users. This means that biased discourses will remain biased if the algorithm does not account for human bias.

Which is why obtaining relevant data is a primary challenge in machine learning. As data is processed, before it can provide input to respective algorithms, it has a significant impact on results which should be achieved. But understanding these results is also a major challenge when understanding the effectiveness of machine learning algorithms. Research needs to be pushed and various machine learning techniques need to be exercised.

Where can machine learning take us?

Work is being done to tackle these biases. In fact, the link between the youth and technology is a factor of this due to their early adoption of tech. As vocal champions of their generation, their ethics stand strong and societal observations are more varied and liberal than any other generation before them. It makes them the voice of whether a brand will sink or swim. In which case, artificial intelligence can be used to target content so long as the creation of content is human generated as brand messages can be easily misconstrued, as is the case with chatbots.

The process of which AI is used needs to be thoroughly understood where regular reviews ensure it is working as required. There needs to be a broad range of input to make the most informed decisions because AI should not be solely relied upon to make decisions — public opinion is always changing, a brand must mould with the times.

More importantly for Brightblue, if AI is used correctly it has the ability to improve effectiveness of marketing activities. The more available data there is the better we can understand customers behaviour. Although this may have the tech industry wired like something out of a Philip K. Dick novel there is still a way to go but the possibilities it seems are likely to startle the modern world should it be successful.

Want to keep yourself up to date with the world of tech and econometrics?

Sign up to our newsletter here:

http://eepurl.com/bcNNyX

See what Brightblue can do for your business

Get in contact