Page 12 - The Indian EYE 031023

OPINION MARCH 10, 2023 | The Indian Eye 12

chatbot ‘Bard’, a conversational AI paganda by various malicious actors.
as a rival to OpenAI’s ChatGPT. A report by CSET and OpenAI dis-
It is said to be capable of respond-
ing to human queries and synthesiz- cusses the three dimensions (actors,
behaviors and content) where lan-
ing information like ChatGPT and guage models and Generative AI
is a lightweight version of Google’s can be used for targeted influence
LaMDA. However, within days of the operations. Considering the pace of
launch, the flaw in Bard was noticed development in this field, these mod-
where the tool made a factual error els are likely to become more usable,
in one of its promotional videos. Fol- efficient and cost-effective with time,
lowing this, Google’s share dropped making it easier for the threat actors
by 9 per cent and the company lost to use them for malicious activities.
around US$ 100 billion in market Generative AI systems have the
value. Google’s Vice President of potential to revolutionize the way we
Search Prabhakar Raghavan asked work and live. Its capability to cater
the trainers and executives to rewrite to diverse audiences with meaning-
Bard’s incorrect responses. Google ful information in a contextualized
is also investing US$ 300 million in manner and provide tailor-made
Anthropic, an AI startup to work responses has brought a significant
in the field of Generative AI. Some breakthrough in technology and how
other generative AI models by Goo- we use it. As these tech companies
gle are MUM, PaLM and MusicLM. dive into the foray of these AI appli-
Microsoft is also said to be in-
vesting billions of dollars in AI and cations and use cases, it is imperative
to study the implications of this tech-
revamping its search engine Bing nology and how it affects society at
and Edge web browser with AI ca- large. The regulation of AI systems
pabilities. It is working in collabo- is still in its infancy, and countries
ration with OpenAI and is looking looking at building their own policies
at integrating ChatGPT into Bing only use the relevant parameters to was claimed as sentient by their en- and regulations can learn from the
and further commercialize its Azure generate a response, making them gineers, and OpenAI’s DALLE-2 positives and negatives of the two
OpenAI service with several AI mod- less computationally difficult. talking gibberish was said to have different models being implemented
els like GPT3.5, Codex and DALL-E The race among the tech giants to created its own language. by the EU and China.
and the soon to be released GPT4. come out with these tools is like the in- Another issue with these Gen- The next wave of innovation in
The big tech companies investing novators’ dilemma to rule the search erative AI systems is that cybercrim-
in Generative AI tools indicate the engine business. The reasons behind inals have started to use these tools Generative AI and LLMs will bring
new use cases and applications in
promise these tools present and the this hurry to come out with such tools to develop malicious codes and tools. other domains with better reliability
profound benefits users experience could be either to take the lead in According to Check Point Research mechanisms. These AI tools certain-
when they come across meaningful this business and vision for the future (CPR), major underground hack- ly have limitless potential, but at the
writings and content that seems to or to collect more data from human ing communities are already using same time, they should not be total-
incur human annotation. These tools users and keep training their mod- OpenAI to create spear-phishing ly relied upon as a replacement for
will bring ease in doing business with els to perform better. Nevertheless, emails, infostealers, encryption tools, human decision-making as they lack
multiple use cases in various sectors adopting these tools will be part of and other fraud activities. The dark emotional intelligence and human
like devising personalized marketing, the businesses soon, but criticism over web is being used by the hackers for intuition and struggle with language
social media and sales content; code their shortcomings will also follow. posting the benefits of malware and nuances and context, with the risk
generation, documentation and con- The challenge with Generative for sharing code (like for generating of biases being introduced at any
tent creation in IT; pulling out data, AI models is to ensure that the gen- stealers) with the help of tools like point in their structural mechanisms.
summarising and drafting of legal erated data is of good quality, bal- ChatGPT. There is no silver bullet solution with
documents; enabling R&D in drug anced, free from potential biases and One of the negative use cases of Generative AI systems, and hence
discovery; providing self-serve func- a good representative of the original Generative AI is spreading disinfor- coordination among stakeholders,
tions in Human Resources (HR) and data. These models present a risk of mation, shaping public perception civil society, government, and other
assisting in content creation for ques- overfitting and generation of unreal- and influence operations. These lan- institutions is needed to manage and
tionnaires and interviews; employ- istic data, which raises ethical con- guage models have the capability to control the risks associated with this
ee optimization through automated cerns related to using such models. automate the creation of misleading technology.
responses, text translation, crafting Last year, Google’s chatbot LaMDA text, audio and videos to spread pro-
presentations and synthesizing infor- Dr Sanur Sharma is Associate Fellow
mation from video meetings; and cre- at Manohar Parrikar Institute for
ating assistants for specific businesses. Defence Studies and Analyses
In the future, these tools are ex-
Niranjan Chandrashekhar Oak is a
pected to generate their own data by Research Analyst at Manohar Parrikar
bootstrapping their own intelligence Institute for Defense Studies and Anal-
and fine-tuning it for better perfor- yses (MP-IDSA), New Delhi
mance. All these tools are based on
an autoregressive transformer mod- This is an abridged version of the arti-
el and are dense, which means that cle that first appeared in the Comments
they use all the parameters (millions/ section of the website (www.idsa.in) of
billions) to produce a response. The Manohar Parrikar Institute for Defense
research in this aspect is now moving Studies and Analyses, New Delhi on
towards designing models that will March 1, 2023

www.TheIndianEYE.com