Among the cutting edge technologies, Artificial Intelligence is perhaps at the peak of its hype cycle. We are witnessing the emergence of next era after the industrial age and the recent Information age. Our lives would be completely transformed in a few decades. AI is not new , it has been evolving. From old terms like Data mining, Expert systems and Data analytics, now everything seems to be Machine learning and Deep learning. There are trends like increasing computing power, falling cost of computing power, availability of huge data platforms and better algorithms have made AI a reality touching our daily lives. Let us look at AI beyond the hype- what is already possible and what are the remaining challenges, with two fields where humans excel- seeing and speaking.
Even a simple task of identifying the object in a picture is quite complicated for a computer. It is like teaching a child with a picture book. Machine learns by itself by ‘seeing’ more and more examples with wide variety, tagged by humans. For example, in the picture below, the three images on the left with captions are from humans and the picture on the right is identified by computer running Deep learning algorithms.
Today such Convolutional neural networks are used – by Facebook to tag you and by Google to search in a picture library with key words given by you. The AI Boom that we see today in Computer vision could be attributed to the events “ImageNet”Challenges. Over the years, algorithms that surpass human ability have been developed by the winners of the challenge. As you can see below, from 2015 , error rates have fallen and the computer performance is better.
Sounds fantastic isn’t it ? Now, what is the hype ?
Though the algorithms show unbelievable error performance, they are not fool proof. By adding noise to the image, the machine could be fooled. These are called “Adversial attacks”. See below the effect of adding noise that makes Panda to be identified as Gibbon with high confidence! But humans will still identify rightly.
Another area in our life, where AI is making large strides is “ Conversation Systems “. Talking with computers has been an old pursuit. It started with Allen Turing , a British mathematician ,crypto analyst and a computer scientist. He published a paper in 1950 titled “ Computing machinery and intelligence “ He opens the paper with the question “ Can machines think ? “, but since “thinking” is too broad and vague, he then proceeds to modify the question as “ Can computers do well in the imitation game ?”. What is the imitation game or the Turing test ? It is a test of a machines ability to exhibit intelligent behaviour like humans in natural language conversation run as a chat. In the conversation between a judge and a computer, executed as a chat on computer screen, if the judge cannot distinguish between a human and a machine, then the machine is supposed to have passed the Turing test.
In 1964, a computer program called “ Eliza , world’s first chatbot passed the Turing test. Its creator Wizenbaum was amazed to find that his secretary was regularly engaged in conversation with Eliza , including personal discussions, and she believed that the computer understood her.
There is an annual contest still running, the Loebner Prize, where chatbots are assessed. The current form of AI solutions is a bot – a software that can do many tasks without human help. They can converse with clients on Web pages smoothly. When you see a pop up window on a web page “ How may I help you ?” – be sure that is a bot and no human is sitting and answering your questions. Bots can be constructed by simple rules and they answer only from a set of questions and limited dictionary. But true AI bots are self-learning and they evolve over time. Siri from Apple and Alexa from Amazon are such wonderful examples. Though you get “ I do not understand that question “ many times, over time they will learn.
Google has made huge progress – Sundar Pichai demonstrated Google assistant in 2018 – booking haircut appointment and a restaurant table ! It is amusing to hear “ Hm Hmm “ at the right place from the machine !! This is way ahead of asking Alexa “ What is weather today ?” Alibaba has published similarly a superior conversation solution recently.
You will see that it is just not text chat but actual natural language conversation. The core technology behind the conversation systems is NLP – Natural language processing. Rather than treating words as sequence of symbols, NLP processes the hierarchical structure of language: words make a phrase, phrases make a sentence and, ultimately, sentences convey ideas. It involves Natural language understanding, deep learning and text to speech! More than all – it understands the context !
Amazing isn’t it !
But have we arrived at the ultimate ? Not yet, there is a long way to go – to understand emotions, to understand the difference between sarcasm and real statement, to understand subtle nuances of individual styles ! That is in summary about the hype and reality.
With latest developments like facial recognition there are serious issues related to personal data. There are many ethical issues being debated. But like any other technology, the usage in our hands. We cannot avoid AI, but let us put it in to right use for us.
About
the author:
SriKrishnan
graduated with a degree in Electronics and Communications Engineering from PSG
College of Technology. He is Senior Vice President Engineering, at Robert Bosch Engineering and Solutions in Bangalore. This article was ortinally published in the PSG Tech 2018 Alumni Congress Souvenir.