Generative AI represents a defining moment in the history of technology and its potential impact that cannot be overstated. Azeem Azhar could not have expressed an anticipated moment like this better in his ‘2021 The Exponential Age’: As new technologies converge and inevitably compound one another’s increases in socially disruptive power, humanity’s ride will only become faster and bumpier from here on out, resulting in an upward “hockey-stick”. Growth trajectory that promises unimaginable benefits while also threatening to tear apart the comparatively staid institutions and social norms we rely upon.

That exponential moment is here, and we can only expect it to multiply further as chips, clouds, and technologies like Generative AI come together. Each will bring with it a level of power never seen before. With its ability to generate novel content, synthetic or new information that did not exist before, generative AI systems has the power to disrupt a range of industries, from art and entertainment to finance and healthcare.

As organizations, policymakers, and investors grapple with the implications of this technology, it is important to understand what generative AI is, how it works, and what are its potential risks and benefits are. In a series of articles, we intend to publish in sprints, we will explore these key standing topics and add more as fit, to bring a comprehensive, practical, and an actionable understanding of this groundbreaking technology and its role in shaping the future.

1. Generative AI options for AI adoption and ongoing development.
2. Actionable and foreseeable uses of Generative AI – from the vantage point of business and technology executives, investors, small businesses and entrepreneurs, and policymakers
3. Risks and impact, and a road map for balanced risk-taking.
4. Generative AI policy guidance and formulation (future edition).
5. The evolving business model and future of workforce.

Among other areas of interest to keep an eye on are generative AI in multi-modal applications and aspects of generative AI that are moving into the realm of human behavior, such as creativity thinking.

The Future of Industries with Generative AI: Deployment Options and Ongoing Development

Generative AI models also known as foundation models, represents a major step forward in our ability to create new content and knowledge. Unlike traditional AI systems that are limited to using pre-existing data to make predictions and decisions, Generative AI can recreate or rehash content, such as images, music, and text, that has never been seen before. This can serve as knowledge synthesized, information augmented or created, or gaps to fill in large datasets and can have profound implications for a wide range of industries, from art and entertainment to science and engineering, and could lead to entirely new forms of creativity and innovation.

Generative AI has the potential to democratize creativity and knowledge creation, allowing people who may not have access to traditional education or resources to contribute to the creation of new ideas and content.

Unlike traditional AI systems that are limited to using pre-existing data to make predictions and decisions, generative AI can produce content, such as images, music, and text, that has never been seen before based on its training and learning experiences. This has profound implications for a wide range of industries, from art and entertainment to science and engineering, and could lead to entirely new forms of creativity and innovation.

The possibilities of foundations model are endless.

Business should evaluate from the context of its construct viability and cost, context of use (core competency or shared service), scale or impact to business and its downside risks.

For most companies, Generative AI is not about the “daily deluge of 10+ productivity tools released” for solopreneurs and small businesses, but rather on how to monetize data, provability of business case and value realization, scale of operations, enablers and tactical execution, talent, and associated compliance and risk mitigation.

Considering the wide array of options and viability available from model selection to product differentiation, a key to business transformation, our upcoming editions will focus on distilling on these matters along with emerging risks & remediation recommendation, in greater detail.

Generative AI can either be deployed out-of-the-box or fine-tuned to perform specific tasks.

The following are common options available for adopting Generative AI use cases:

OPTION 1 – Usability at the individual level: For example: $20 monthly subscription to ChatGPT or GPT-4

OPTION 2 – For light-weight productivity tools and similar applications: Through the use of open-source APIs available for businesses that want to develop applications on top of its foundational models. For example: OpenAI APIs are leveraged to develop applications typically for marketing, productivity gains, and for creating assistants aligned with specific business needs.

OPTION 3 – Specialized vertically aligned AI engines: Through the use of open-source API on top of its foundation models. This option typically does not differentiate
or support personalization.
If that is not a concern, foundation model like Open AI or Bard AI should support business needs. For example: Answering complex questions, pulling from vast amounts of legal or financial documentation, and drafting and reviewing annual reports, earning reports, etc.

OPTION 4 – Use of off-the-shelf open source large language model (LLM): A cheap-to-build LLM model that supports AI democratization for enterprises wanting to build their own generative AI language models. For example: Databricks introduced Dolly, affordable model, that exhibits an instruction-following capabilities similar to ChatGPT. Trained on 6 billion parameters as opposed to 175 billion parameters in GPT-4.  The LLM is cloned from the Alpaca open model built by Stanford.

OPTION 5 – As a service Foundation model: Provides ability differentiate on specialized use cases or to gain competitive advantage. Using services, companies have an option to build (personalized database) on high-quality pre-trained models specifically built and trained on specialized industry use cases, where developers may further customize or fine-tune. For example: NVDIA has over 50 specific use cases to explore across NLP, Speech AI, Computer Vision, Healthcare, Cybersecurity, Art and Creative workflow, etc.
Scalable pretrained AI models can be applied across industries. Further customization and fine-tuning can lead to infinite possibilities for use cases.

Additional Notes on Options:

Use cases built on option 1 and 2 are released on model hubs or as open source, enabling fine-tuning pretrained or foundation AI models. Subsequent training will be necessary to improve their efficiency and accuracy and expand model repositories.

For options 2 and 3, there are myriad of open-source model options available on hubs like OpenAI, GitHub, Hugging Face and others that offer plenty of choices to companies and developers alike.

NVIDIA’s as a service solutions foundation model is a family of cloud services with which enterprises will be able to build their own LLMs and run them at scale, calling them from enterprise applications.

In NVIDIA’s recent developer’s conference, NVIDIA outlined 3 services launched on limited access. Each include a pretrained model, data processing frameworks (providing raw data for differentiations), ability to couple personalization databases, inference engines and APIs to access the service.

Examples of these include:

• NeMo for generating text
• Picasso for visuals – So to elaborate, use of in the recent Adobe Firefly (Adobe debuting its beta-version to generate images & videos from text) and soon to be integrated into Abode Photoshop, Illustrator and Express, and NVIDIA’s own Picasso/Imagery systems for generating images, videos and 3D apps
• BioNeMo for molecular structures

Every entity building open-source foundations or specialized pre-trained models, we stress on the importance of using ethical data that is transparent, explainable, and obtained with consent and without bias in order to train pre-trained models.

As we look through each option, the risks associated with the model remains mostly the same. All options require diverse and multi-stakeholder disciplinary to examine and mitigate risk from concept to deployment. The data inputs, pipeline data, technology and data capabilities stack, and outcomes should be part of a robust AI governance framework together with data governance, risk management practices, and regulatory compliance.

Actionable and Foreseeable Uses of Generative AI

A view from the perspective of business and technology executives, investors, SMBs, entrepreneurs, and policy makers.

Generative AI has the potential to impact a wide range of constituencies, including investors, mid to large enterprises, small businesses and entrepreneurs to lawmakers and regulators.

From an investor’s point of view –

Generative AI presents an opportunity to invest in a cutting-edge technology with a vast potential market.

As generative AI continues to advance, the tools used to build and manage these systems are becoming increasingly important.

At the bottom of the stack are the compute resources, which must be powerful enough to train and run these complex models. On top of that are the machine learning frameworks, such as TensorFlow and PyTorch, which provide the building blocks for developing generative AI applications.

In addition to the frameworks themselves, InsightPartners provide investor’s point of view on tools. One of them being the management and safety, compliance, and security concerns and requirements surrounding large language models as critically important.

As Generative AI becomes more prevalent in business and society, it is important for companies to carefully consider their choices in the evolving AI stack to ensure they are able to effectively build, manage, and secure these systems.

From business and technology executive point of view –

we are witnessing a transformative moment in the business world as companies across various industries are in awe with the potential of generative AI. We want to underscore ‘the transformative moment from the aspect of exploratory stages’ where most companies are still at. Businesses are taking stock of their data and assessing their technology capabilities, evaluate options (see our topic on ‘Deployment Options and Ongoing Advancements’) drive differentiation and maximize the upside risks and usability. Let’s take a closer look at some practical business uses of generative AI by industry and use case.

Healthcare:

• Personalized treatment plans generated from patient data
• Medical image interpretation and analysis for more accurate diagnoses
• Drug discovery and development through generative models

Financial Services:

• Fraud detection and prevention through analysis of large amounts of data
• Portfolio optimization through generative models that simulate market scenarios
• Customer segmentation for more targeted marketing and product offerings

Legal:

• Contract analysis and review through natural language processing
• E-discovery for more efficient document analysis and review
• Prediction of case outcomes through analysis of past cases and legal precedents

Retail:

• Recommendation engines for personalized product offerings and targeted marketing
• Inventory management and optimization through analysis of sales data and consumer behavior
• Virtual try-on tools for online shopping

Supply Chain:

• Predictive maintenance through analysis of sensor data from equipment and machinery
• Optimized route planning and logistics through analysis of transportation data
• Demand forecasting for more efficient production and inventory management

Manufacturing:

• Predictive maintenance through analysis of sensor data from equipment and machinery
• Product design and optimization through generative models
• Quality control through analysis of manufacturing data for defects and anomalies

These are just a few examples, but the potential uses of generative AI in various industries are vast and constantly evolving.

From small businesses and entrepreneurs point of view –

Generative AI offers a cost-effective solution to synthesize high-quality content, providing fit-for use and scale opportunities in the market. The increasing availability of AI tools tailored to support smaller enterprises is staggering, and the market is rapidly evolving. As the industry continues to mature, we anticipate companies will explore various applications and carve out niches within their respective markets. We will continue to monitor these trends and provide updates in our upcoming editions.

From Lawmakers and regulators point of view –

Lawmakers have an interest in ensuring that the technology is used ethically and responsibly, and may focus on issues such as privacy, security, and bias and safety, and myriad of other constraints inherent to this technology.

With the AI market catching the attention of regulators, we are now faced with the sudden emergence of Generative AI, which faces several policy and regulatory challenges. With several AI enactments in the pipeline such as the EU AI Act along with the 2021 AI Bill of Rights that finally arrived for public scrutiny will further face critical judgements against Generative AI.

AI Bill of Rights: In February 2021, the US Federal Trade Commission (FTC) released a set of guidelines for companies developing or using AI technologies. These guidelines, known as the AI Bill of Rights, aim to promote transparency and accountability in AI systems. The guidelines include principles such as explaining the use and purpose of AI, ensuring accuracy and fairness of AI systems, and protecting consumer privacy and security.

EU AI Act: In April 2021, the European Commission proposed a set of regulations for AI, known as the EU AI Act. The act aims to promote trustworthy AI that is safe, respects privacy, and is transparent. The regulations cover a range of AI applications, from high-risk systems such as biometric identification and critical infrastructure to less risky applications such as chatbots and recommendation systems.

State Regulations: Several US states are also taking steps to regulate AI tools. For example, in 2021, Massachusetts proposed a law that would require companies to conduct impact assessments for AI systems used in state agencies. In New York, a bill was introduced to establish a task force to study the impact of AI on employment and the economy. Other states such as California, Washington, and Illinois have also enacted laws related to AI ethics, data privacy, and bias.

While the EU AI Act does not specifically target generative AI models like ChatGPT and family of Generative AI models, some lawmakers have proposed designating them as “high-risk” under the legislation, which focuses on regulating the use of AI rather than specific technologies. However, the use of generative AI models presents clear challenges to data management not only under the EU AI Act but also under GDPR, which enshrines seven principles on personal data processing.

The shift in data usage from human-generated to AI-generated content raises concerns about data sources, privacy and protection, and policymakers must consider how to address these challenges in regulating the use of generative AI.

Overall, the interests and benefits of each constituency are diverse, and will continue to evolve as the technology develops and becomes more widespread.

Risks and Impacts

In GPT-4’s own words when asked of its risks and limitations:

As an AI language model, I have some limitations that are important to keep in mind:

1. Bias: Like any AI system, I can reflect the biases that exist in the data that I was trained on. This means that my responses may not always be objective or unbiased, and may reflect societal or cultural biases. It’s important to be aware of this and to take steps to mitigate bias when using AI language models.
2. Context: While I’m designed to understand and respond to a wide range of prompts and questions, I still have limitations when it comes to understanding context. This means that my responses may sometimes be incomplete or inaccurate, particularly when dealing with complex or nuanced topics.
3. Creativity: While I can generate responses that are often surprising or unexpected, I’m still limited in my ability to generate truly original or creative content. This is because I’m essentially a pattern-recognition system, and my responses are based on the patterns I’ve learned from the data I was trained on.
4. Ethics: As an AI language model, I’m not capable of making ethical or moral judgments. This means that I may provide responses that are technically accurate but may not always be ethical or appropriate in a given situation. It’s important for users to exercise judgment and critical thinking when using AI language models to ensure that they are making ethical and responsible decisions.

Overall, while I’m capable of performing a wide range of language tasks, I have limitations that are important to keep in mind. It’s important to use AI language models responsibly and to be aware of their limitations in order to ensure that they are used in a way that is beneficial for everyone.

As with any new technology, generative AI also comes with a host of uncertainties and challenges that applies to any of the options laid out in ‘Deployment Options and Future Developments’ section.

One of the most pressing concerns is the potential for bias or other compliance setbacks in the generated content particularly in its ability to comply with the GDPR privacy principles (as things stand).

In traditional models and going into as far as supervised models, the outcomes are focused on a problem to be solved. While it carries risks that are of common denominator to generative or foundation model, the threat landscape was slightly behind on the emergent nature of uncertainties found in Generative AI.

Unlike traditional models where the outcome is more one-dimensional, Generative AI can produce outputs that are complex and compounded, difficult to interpret or analyze, leading to potential ethical or legal issues.

This means organizations working with generative AI need to conduct careful due diligence and take additional measures to ensure that the outputs are fair, interpretable for validations, transparency in the data and model construct, reliability of its adaptive nature and comply with legal and ethical standards. This can further exacerbate the threat landscape especially when introduced with other potential security vulnerabilities and exploits such as adversarial and data poisoning attacks.

Generative AI is a league of its own when it comes to risk management and growing pains for regulators.

In the upcoming edition, we will showcase the best practices and considerations in the management of risks associated with the use of Generative AI.

The Evolving Business Model and Future of Workforce

“I can’t emphasize this enough: Open AI would not exist without NVIDIA,” mentioned Jim Cramer, host of CNBC’s Mad Money. In his recent Cramer’s interview with NVIDIA CEO Jensen Haung on advancements with AI accelerated computing during the NVIDIA annual developer’s and AI conference (GTC), the CEO gives a clear indication of expected future transformation changes across business models – in the way we manage and operate companies.  These radical market indicators have not escaped the attention of Wall Street, which has come to recognize the power of AI.

NVIDIA’s CEO has touted Generative AI as the iPhone moment for AI and calls it a big deal – the recent changes have created a new computing model – the way you program is different, the way you use is it different and the reach and accessibility is different.

In simple terms he describes Generative AI as an instrument that can write software for you, write stories and poetry for you, do an art for you, condense a contract for you, it can be precise or imprecise, the computer figures it out and does it for you. The solution can be deployed in many different ways as demonstrated by Microsoft and Google, by integrating with the most pervasive office automation software in the world.

There is no doubt that these indicators are the result of a brand-new computing model that has the potential to unleash existing business models by radically transforming the way companies capture, create, and deliver value to customers. Generative AI such a large language models is now proving to as an application that is easy to use and effective when you tell it to do something. The adaptability aspect of this was demonstrated by the 100 million+ users on ChatGPT emphasizing the ease of user adoption and change management. Per NVIDIA CEO, Generative AI is a true Apple moment for AI.

As companies explore transformational ways to unlock value and maximize performance, it will be the responsibility of companies adopting these models to implement appropriate guardrails. This will ensure bias and safety checks, protect privacy and demonstrate ethical behavior and secure the environments these systems are embedded in, as these are much debated and concerning topics of large language models.

Although we touch as its ease of usability, integrating generative AI into existing business models can be challenging. A road with many pathways to a destination and still in its exploratory stage. It requires careful planning and a willingness to change existing business processes and workflows. Additionally, it may require the development of new skills and the recruitment of new talent. For example, organizations may need to invest in data scientists or AI engineers to develop and manage the generative AI systems.

Well before the Generative AI hype, Garter predicted AI talent shortage to intensify the market demand and now, this may have compounded effect with the emergence of Generative AI.

Another challenge is the potential for bias in the generative AI systems. Since the output is generated based on existing data and learned experiences, it may reflect any biases present in that data. This should be of particular concern, especially in the absence of an effective AI governance oversight between senior management and teams on-the-ground where biased AI systems could perpetuate existing inequalities. To address this, organizations will need to carefully plan, design, develop and train the model, evaluate their generative AI systems to ensure that they are fair and unbiased and impacts of all residual risks are clearly understood before deployment.

Overall, while generative AI offers significant benefits to businesses, it also presents significant challenges. Successful integration requires a careful balance between technological capabilities and associated risks unique to these models, as well as a willingness to adapt existing business models and processes to take full advantage of this transformative technology.

Closing Remarks

In conclusion, generative AI is a game-changer that has the potential to revolutionize numerous industries and disrupt the traditional ways of doing business. From its ability to generate lifelike images, music, and text to its power to drive innovation and create new opportunities, the possibilities are endless. Even so, it is not without its uncertainties and challenges, and stakeholders have the responsibility to ensure that technology and data are used responsibly.

In the next edition, we will delve into some practical applications of generative AI and explore how businesses can leverage its power to gain a competitive edge. We want to include new topics such as Policy Guidance and Development, Risk Management unique to Generative AI, effective Governance. Stay tuned for more exciting insights on this defining moment in history.