While driving a car,
the GPS system’s human voice guides you to your destination. Virtual
assistants, Alexa and Siri, answer a wide range of
questions, play music, and control smart home devices. In the
entertainment industry, special effects turn ordinary mortals into super-beings
on the screen; fake war scenes and action sequences appear genuine enough to
keep audiences riveted to their seats. A politician’s 3-D holograms appear in
several places simultaneously during election campaigns. Fake narratives and
doctored images cause violence and polarization, causing death, destruction and
misery. This is the power of artificial intelligence.
Indeed, artificial
intelligence has taken the world by storm. But there are conflicting opinions on
its positive and negative effects on the world. To see things in perspective,
let us take a brief look at its history.
AI, or artificial
intelligence, has been around for decades. In 1932, Georges Artsruni created
a device that he called a “mechanical brain”. It could translate languages. He
patented it in France in 1933. In 1937, he built its more advanced version.
Artificial Intelligence
differs from Generative Artificial Intelligence.
1.
Artificial
Intelligence analyzes data and provides conclusions or results which help in
decision-making. For example, it can detect frauds by detecting patterns, evaluate
and classify data relating to a crime.
2.
Generative
Artificial Intelligence helps produce new content, chat responses, designs,
synthetic data, and shallow as well as deep fakes.
Origins and Spread of
AI
In 1957, linguist Noam
Chomsky published “Syntactic Structures” that explained the parsing and
generating natural language sentences. This slim volume had a major impact
on the study of knowledge, mind and mental processes and became a milestone in
the development of cognitive science.
In 1966, Professor
Joseph Weizenbaum of the MIT created a chatbot named Eliza. It simulated
conversations with a psychotherapist. However, it had several shortcomings, such
as a limited vocabulary, lack of context and overreliance on patterns, etc.
In 1968, Professor
Terry Allen Winograd created the first ever multimodal AI that could understand
natural language. The program accepted commands such as, “Find a block which is
taller than the one you are holding and put it into the box”. The program could
also respond verbally, for example, “I do not know which block you mean.”
In 1980, Michael Toy
and Glenn Wichman developed a Unix based game called Rogue, which dynamically
generated new game levels. Later, Jon Lane joined them to create a commercial
version of the game.
In 1985, Judea Pearl
developed Bayesian networks to define complex probability problems. This
revolutionized AI, making it an important tool in natural sciences and
engineering
In 1986, Michael Irwin
Jordan laid the foundation of Recurrent Neural Networks. This helped perform such
complex work as language translation, natural language processing, speech
recognition, and image captioning; popular applications, such as Siri, voice
search, and Google Translate, are cogent examples.
In 1989, Yann LeCun,
Yoshua Bengio and Patrick Haffner demonstrated how Convolutional Neural
Networks or CNN can be used for recognizing objects with gradient-based
learning. These networks help recognize multiple objects without requiring
explicit segmentation of the objects from their surroundings. Later, they
developed Graph Transformer Network model, which extended the applicability of
gradient-based learning to systems that used graphs to represent features,
objects, and their combinations.
In 2006, data scientist
Fei-Fei Li came up with the idea of ImageNet at the University of Illinois.
Most AI researchers thought algorithms were more important than the data
itself. However, Li proved that vast amounts of real-world data would make
algorithms more accurate.
Today, thanks to
powerful graphic cards and computers, deep learning seems the most promising
for several applications (including voice or image recognition). Experiments
conducted at Microsoft, Google and IBM showed that this type of learning halved
the error rates for speech recognition. It enabled considerable progress in
text recognition.
ChatGPT, Dall-E and
Bard can produce a wide spectrum of content, ranging from poetry to art and
medical reports. It has become possible to automate writing of content and
respond to specific technical queries. Creation of realistic representations of
people and objects has become workable. AI can provide coherence to complex
data and narratives. In fact, the role of AI in everyday human activities has
just begun. It is bound to expand and enter the world of creativity, science,
technology, industrial production, business, travel and entertainment etc in a
big way, thus creating unimaginable problems around the world.
Challenges Posed by AI
Already, the AI driven
DeepFake has demonstrated dangers of its unrestrained use. There have been many
instances of faces of celebrities replacing actual actors in pornographic movies.
Doctored videos of politicians for malicious use have become quite common.
Since fake audiovisuals of real persons can be created, the potential for
serious mischief has become dangerously deep. International incidents – say
between Ukraine and Russia, or China and India – can be triggered with the help
of faked alarms. The “bombing of Pentagon” is a telling instance.
There are other
problems too. These may not appear to be as destructive and dangerous, but can
harm the society immensely. Plagiarism of art and literature is increasing. In
January 2023, three popular art generating platforms, viz., DreamUp, Midjourney
and Stable, faced copyright charges from US artists: Sarah Anderson, Kelly
McKernan, and Karla Ortiz. A platform named Stability AI had to face a lawsuit
from Getty Images for plagiarizing millions of images from its stock
collection. Authors of research papers, nonfiction and fiction works feel
threatened by AI driven platforms like ChatGPT, GPT4, etc. When original and AI
generated content can be similar and AI platforms can copy anything, there is
an urgent need for putting a strong legal structure in place for preventing
firms from using the work of writers, artists, singers to train their AI models
without the creators’ consent.
Regulating the AI
The
demand for regulating the use of original and copyrighted content by generative
AI platforms is rising all over the world. The solution lies in AI itself.
Significant progress has been made to identify fake videos. Artificial
intelligence already helps to spot fake videos and image-based disinformation
attempts. India’s AltNews has done a commendable job in this respect. Detection
systems are being further developed to flag fakes whenever they appear. A
blockchain online ledger system could hold a tamper-proof record of videos,
pictures and audio, so their origins and any manipulations can be checked.
On the regulatory
front, the European Union is reportedly taking steps to come up with
comprehensive laws and regulations for reining in the untrammeled infringement
of patents and copyrights. India has been slow, but surely it will have to put
its act together.
No creation of artificial
intelligence will ever overwhelm human genius.
PostScript: On a lighter note,
recently in the Indian state of Telangana, aspiring civil servants used ChatGPT
to pass the exams. According to a report by Times of India, an aspirant used
ChatGPT to answer the leaked question papers for the assistant executive
engineer (AEE) and divisional accounts officer (DAO) exams. With the aid of
Bluetooth earbuds, he discreetly transmitted the answers to fellow aspirants
during the examination.
AI cannot match the human
ingenuity.
No comments:
Post a Comment