GPT & The End Of The Middle Class

After the previous article (ChatGPT for Non-IT Researchers), there were several requests for a landscape overview?of the GPT phenomenon addressing two key aspects:

• What are the opportunities to utilize GPT for practical applications? What is the monetization playbook?
• What are the socio-economic projections of GPT deployment at scale? What will its economic impact be beyond the 10-year horizon?

In this article, we address these questions in reverse order, first the big picture, then the actionable takeaways.

• Work to make it happen, watch while it happens, or wonder what happened!
• How GPT is about to revolutionize the economy
• Impact on the labor market
• Whose jobs are relatively safe for now?
• Impact on public policy
• Competitive dynamics of the GPT marketplace
• Business implications of the GPT revolution: What should companies do?
• Scoping the gold rush
• Business case definition for GPT Applications
• "You are terminated" or "I will be back"?

Work to make it happen, watch while it happens, or wonder what happened!

In this, the century of future shock, forward-thinking organizations must always be alert for new technologies that might revolutionize or disrupt their environment or operational model. AI is changing the world as we know it, and if you don't keep up, you'll be shocked when you look around to a world that has passed you by!

Will GPT make the already troubling income and wealth inequality even worse? Or could it help? Though it will provide a boost to productivity, will the benefits accrue only to a few tech millionaires or also to the middle and lower class? Who will control the future of this amazing technology? New large language models will transform many jobs. Will it lead to widespread prosperity or spell the end of the middle class?

App developers, venture-backed startups, and some of the world’s largest corporations are all scrambling to make sense of the sensational text-generating bot released by OpenAI in November 2022. You can practically hear the shrieks from corner offices around the world: “What is our ChatGPT play? How do we make money off this?”

How GPT is about to revolutionize the economy

AI tools help the least skilled and least accomplished workers the most, decreasing the performance gap between employees. In other words, the poor performers get much better; while the good performers simply get a little faster.

GPT and other generative AIs could, in the jargon of economists, “upskill” people who are having trouble finding work. Workers displaced from office and manufacturing in one domain might be able to leverage generative AI as a practical tool to broaden their expertise and acquire the specialized skills required in other areas, such as health care or teaching, where there are plenty of jobs, revitalizing the workforce.

Less-skilled workers will increasingly be able to take on what is now thought of as knowledge work, and at the same time, the most talented knowledge workers will radically scale up their existing advantages over everyone else. This effectively will eliminate the “middle class” in knowledge work.

Automating creative and reasoning tasks previously thought to be solely in the realm of human ability will take robotic process automation to the next level, eliminating some white collar jobs of a less-creative nature. Automating relatively routine workflows will depress job prospects for clerical office workers.

Companies will use AI to destroy what once looked like well-paying automation-proof jobs, and lay off millions of “routinely creative” white collar workers. Those laid off will be forced to shift to lower-paying service employment.

Having more education and expertise no longer will serve as a moat for job security unless the contributions required by the job are of a particularly creative nature. Ten years from now, AI will start eliminating those jobs which are well-paid but non-creative.

Remuneration for performing boring office tasks / routine white-collar drudgery will drop sharply, putting at risk the very existence of such employment opportunities.

However, labor market supply and demand will depress clerical?salaries sufficiently for human labor to remain a competitive alternative to AI-powered robotic process automation. With massive layoffs reconciling people to a universal salary reset, the water eventually will find its own level and a new normal will emerge and stabilize.

There will be significant impact causing structural dislocation in developed and wealthy countries. Poor and less developed countries still exist largely in the world of a century ago, so, ironically enough, they will experience a much smoother transition to the new normal.

Impact on the labor market

Roles heavily reliant on subject matter expertise and critical thinking skills like pharmacists, lawyers and astronomers are an exception, as they have a negative correlation with LLM exposure, and hence may see less impact of an adverse nature.

On the other hand, routine programming and writing skills are positively associated with LLM exposure. Future tools based in part on GPT will write essays and marketing pitches, program boilerplate portions of computer code, and extract insights from financial reports.

ChatGPT based tools might reduce by half the amount of time needed to complete various tasks of these professions, though not fully automate jobs to be done, resulting in merging of roles and reduction in staffing.?

Highly creative / skilled jobs will remain safe.

• The min-bar of creativity to get hired and to stay employed at high compensation levels will rise higher & higher as the performance of AI improves over the years.
• Research and creativity jobs will continue to require human involvement in the foreseeable future.

Some affected professions could see great disruption:

• Interpreters and translators, poets, lyricists and creative writers, authors and workers in publishing industries
• Public relations specialists, survey researchers, news analysts, reporters and journalists, broadcasters
• Legal secretaries, administrative assistants, clinical data managers, climate policy analysts, wholesalers
• Court reporters, simultaneous captioners, proofreaders, copymarkers, correspondence clerks
• Cost / tax accountants, estimators / auditors, quantitative analysts / mathematicians, financial security / commodity contract / credit agents
• Database administrators, web & digital interface developers, blockchain engineers and workers in data processing hosting

Some professions will undergo role revision / consolidation due to partial elimination of tasks / jobs to be done:

• Mid-managers, Teachers, Musicians, Artists, Graphic designers, Animal scientists, Marketing strategists, Financial / investment managers
• Insurance appraisers & damage assessors, Retailers & Salespersons, Nurses, Ambulance workers, Tourist guides, Repairers
• Coal / oil / gas workers, Postal workers, Couriers, Launderers, Drivers, Recyclers, Hoteliers, Sports & Entertainment support workers

Whose jobs are relatively safe for now?

Low-paid manual laborers are less likely to be adversely impacted to a significant extent:

• Agriculture, forestry, logging & wood workers, miners, heavy industry & civil construction workers
• Food preparers / servers, dishwashers / floor cleaners, orderlies & warehousing / storage handlers
• Furniture makers, leather, plastic, rubber, paper, textile & metal craftsmen, social assistance coordinators?
• Customer service representatives, bank tellers, electricians, barbers, medical assistants

Unstructured real-world hands-on jobs will remain safe.

• Hands-on blue-collar work in unstructured real-world environments that cannot be substituted by factory-style automated robots will continue to require human involvement in the foreseeable future.
• While agriculture and fisheries will see growth in AI-powered automation, animal husbandry, pet care and work animal management will continue to require human involvement in the foreseeable future.

Job counts in human-to-human in-person services providing affection / care, amusement / entertainment, bliss / delight, enjoyment / pleasure, happiness / satisfaction, excitement / thrill, luxury / style, confidence / trust, peace / safety, etc. will begin expanding slowly until over a few decades they take pole position and start to dominate the economy for the masses.

Universal Basic Income will not materialize but such human-to-human in-person services will evolve to play the same role, with wealthy people maintaining a retinue of personal-service providers whether employees or freelancers.

Tech-savvy companies could quickly become so much more productive that they dominate their workplaces and their sectors, leading to corporate consolidations and the emergence of monopolies. The large computational costs required to build and to run LLM software create a barrier to entry for anyone looking to compete.

Labor unions depend for their bargaining power on the fact that their jobs are essential for the organizations in which they serve. AI is headed inexorably on an inevitable path of destroying such jobs deemed “essential”, neutralizing the negotiating power of unions, and forcing workers into service economy jobs that constitute discretionary consumption from the employer’s perspective.

Thus, while overall productivity and gross national product will increase, so will income inequality and political polarization, continuously squeezing the middle class of society, promoting a minority to the upper class and demoting the majority to the lower class, until a bimodally distributed population with practically no middle class is left.

Impact on public policy

Stagnant wages and worsening income & wealth inequality already has driven political realignment of the blue collar demographic and the emergence of populist protectionist leaders. Now GPT and its ilk promises to?expand the vote-bank of protectionist populism.

Publicly funded national and intergovernmental research, equipped with the huge computing power needed to run the models and the scientific expertise to further develop the technology, could build non-private AI models available as AI superhighways for democratizing access and bring diversity to innovation priorities.

Well-funded bad actors could build DarkAI to breach security & privacy. Unlike Responsible AI, DarkAI additionally could be trained on dark web data, and omit ethical curation filtering pipelines. This will have global security and law & order implications.

The Microsoft-OpenAI alliance will force Alphabet, Amazon, Apple, Meta and other large western world players to bring their own GPT++ offerings to market. Chinese companies will go head to head with the West, leveraging regulatory protections as a competitive advantage.

Vertical integration by strategic alliances will create multinational oligopolistic cartels led by platform companies that control supply and prices across plug-in ancillaries and client industries worldwide.

Governments will respond by enacting compliance laws citing national security, economic sovereignty, restrictive trade practices, labor rights and consumer protection.

Competitive dynamics of the GPT marketplace

OpenAI GPT’s commercial strategy is structured around offering gated API access to selected partners (while withholding access to the underlying GPT model and its trained weights), accelerating go-to-market of startups building business applications.

API access is a much friendlier paradigm for small teams and product-driven startups than building their own LLMs. It turns AI from an annotation / training problem into a meta-coding problem with English as a programming language (prompt engineering).

Small players and new market entrants can create MVPs without investing in much training data or infrastructure. Building AI-powered product features becomes as simple as asking GPT to answer the right questions.

Instead of fine-tuning models with huge training data, one designs and fine-tunes questions in natural language. The approach yields useful results for a large-enough range of problems to trigger a gold rush of ventures over the coming decade.

There will be an explosion of startups building MVPs on OpenAI’s GPT API. MVPs that deliver acceptable performance will start flooding the market. Current market leaders (like Grammarly in the NLP area) will face low-cost competitors.

API-dependent startups may lack domain expertise and marketing muscle to scale beyond the MVP, eventually driving market consolidation. But the cut-throat pricing war for market share will drive non-AI providers into the ground.

Business implications of the GPT revolution: What should companies do?

Generative AI products have vast potential for practical applications in business. Their impact will span every industry including law, finance, marketing, retail, energy, construction, fashion, entertainment, manufacturing and distribution.

Designing innovative new products and creating innovative solutions that meet the specific needs of customers will make GPT a game-changer. In conjunction with complementary products, it can be used to create entirely new content beyond just language and multimedia, such as software code, 3D models and intelligent voice assistants.

GPT can be applied to a wide range of business cases, such as marketing communications, software development, employee training and research. By automating tasks and optimizing business processes through AI-powered digital transformation, GPT will help re-imagine the way different aspects of the businesses operate in multiple contexts.

The nature of daily work will change, as AI-powered “copilots” get embedded and tightly integrated into workplace systems and become standard operating procedures. “AI Certified” will become a recruitment & promotion criterion.

MaaS (Model as a Service) may prove to be a viable business option, because it positions extremely large task-agnostic models as OpEx instead of CapEx. Client industries have no option but to adopt AI to survive the profitability race. Defensive plays like the so-called ChatGPT Detectors are doomed to fail. Using AI to detect AI only makes AI even smarter in the next iteration. An arms race of adversarial discriminators just provides labelled data for fine tuning.

Scoping the gold rush

Learning best practices and practical experience will provide a clear understanding of how this technology can be leveraged to drive business growth and innovation along hitherto inconceivable lines, like (say) automatic generation of business plans and customized software. However, the core remains sound business common sense.

Fitness for purpose: Like every other technology, GPT is one of many tools in the product toolbox. We need to move past scientific achievements and think for what type of applications LLMs such as GPT are best suited.

Though GPT LLMs can perform many tasks, they’re not necessarily the ideal solution or even the best tools for those tasks. How do you decide where to use LLMs versus an alternate solution? Considerations include:

• Customer-facing / internal-use scenarios that require so much sophistication
• Need for generating content / data / programs as part of business process
• Need for accuracy of responses with factual up-to-date info and correct of math, without hallucination and “confident falsehoods”
• Need for non-European languages and niche domains not covered by the current LLM version

Trustworthiness: One guiding principle to determine whether LLMs are fit for an application is whether their mechanism is coherent with the structure of the task. Even if LLMs solve or successfully perform on some examples of the task, it does not mean that they are reliable in that field more generally.

Next-token prediction is an excellent solution for some tasks, such as text generation. However, LLMs are known to hallucinate, making up things that are not correct, so cannot be trusted as sources of knowledge, especially in domains like health, legal advice, science or software coding.

Prompt engineering methods require that you know enough about that domain to be able to provide the extra knowledge for vetting. Blindly trusting the language model can result in output which is unusable due to poor quality and / or due to creating business risk.

In practice, only trust GPT when:

• You are knowledgeable enough about the broad area, have subject matter expertise in the domain and already know the specific topic.
• You are competent to fully review and verify the output.

Commercial Viability: Consider both conventional business factors and product market fit:

• Time to market: Rapid entry / offering launch for mind share and usage share
• Pay-off period: Short horizon to obsolescense due to too rapid evolution of technology and markets
• Cannibalization: Disruptive impact on revenue and RoI of existing lines of business
• Value engineering: Price-utility trade-off for customers of alternative product feature bundles
• Niche targeting: Custom solutions for overlooked segments with small but significant user base
• Demand: Actual market validation of willingness to pay the required rate of return

Performance & Adaptability: Every token GPT generates comes at an enormous computing cost. From a product perspective, the application may not need that kind of heavy computation. In many?cases, even if GPT provides a perfect answer, there might be much simpler, faster, and less costly solutions. Consider aspects like:

• Efficiency: Training time (agility) and Inference time (latency)
• Sizing: Problem size fitting within model’s inferencing capacity limits on input length, context length, query rate and response latency (or feasibility of workaround methods for handling larger problems)
• Configurability: Composable library of small LMs rather than single large LM
• Customizability: Ability to fine-tune on task-specific dataset for higher accuracy

Governance: This covers a spectrum of factors:

• Lack of institutional knowledge management.
• Risk to business due to safety / malpractice, security / privacy, copyright / plagiarism, bias / discrimination, abuse / slander, abetment / incitement, age / sex / drugs / weapons related illegality / criminality, irrelevance / nonsense, controversy / ethics and PR damaging to reputation / goodwill.
• Monitoring of LLM based solutions. Unclear how to put guardrails on chats going offtrack, how to detect hallunications by automated validation scripts, how to prevent malicious users from manipulating session results.
• Unknown nature of data used for pre-training.

Talent: Even while GPT forebodes job losses, it also renews the eternal war for talent:

• Training for traditional developers on generative pre-trained transformer technology and ML training techniques and the complexity of ML engineering / MLops.
• Extreme shortage of expertise: business analysts & technical program managers as well as of data scientists & ML engineers with relevant skills.
• Consulting & IT services firms are quoting exorbitant rates. No single firm offers an end-to-end full service portfolio, so need to work with multiple AI firms.

Costing: Capex is needed to build, maintain & enhance, Opex is needed to deploy and run LLMs. Retraining an open-source LLM with private data and hosting on cloud is unaffordable for mid-tier companies. LLM APIs hosted by OpenAI and its competitors are too expensive.

Business case definition for GPT Applications

Brainstorm use cases across functions and sub-functions in three categories:

• How can we run our company more efficiently?
• How can we support customers more effectively?
• How can we build the brand more creatively?

Be open to the potentialities for GPT-powered innovation in all sectors.

• Finance & Legal: How can we use GPT technology to leapfrog traditional banking? Can we leverage it in apps for payment, insurance, contracts?
• Media: How can we use GPT technology to transform leisure time? Can we leverage it in apps for video sharing, livestreaming, esports and gaming, advertising?
• Life Sciences & Health: How can we use GPT technology to accelerate medical development? Can we leverage it in apps for biological research, data-driven health solutions?
• Mobility: How can we use GPT technology to offer mobility as a service? Can we leverage it in apps for ridesharing, autonomous vehicles, home delivery?
• Agro & Food: How can we use GPT technology to improve agriculture, horticulture, food production & distribution? Can we leverage it in apps for marketing urban farming of alternative protein, promoting rural and traditional agricultural industries?
• Climate & CleanTech: How can we use GPT technology to tackle the climate crisis? Can we leverage it in apps for championing the reduction of greenhouse gas emissions to meet climate commitments, evangelizing clean transport and renewable energy?

There is no sector, whether Ecommerce or Industrial process control or Space or Software & IT, where one cannot find some or the other organizational sub-function in which to use GPT technology for overcoming business, commerce, engineering or scientific challenges. We just need to question our hard-held assumptions about "how things normally are done around here" while retaining pragmatism about scoping and viability.

• Decide which part of the current workflows will be replaced / augmented.
• Address potential ethical and business concerns. For some use cases, it is a show-stopper.
• Cost, LLM inference speed / latency, output variability and infrastructure sophistication make certain use cases impractical or impossible.
• Detail out the business processes to sub-process level and selectively automate the easiest sub-processes first.
• Do not assume the user interface has to be chatbot-like. Business applications can use GPT API on the backend for information retrieval, build value-adding modules on top of that and keep the customer relationship in their own (possibly non-chat) UX.

As an example of how detailing workflows to sub-processes level can expose viable opportunities, consider the use of GPT in educational content creation and presentation. Instead of simply asking GPT to create educational content on a topic, ask what steps a teacher goes through over the educational lifecycle. It will become apparent immediately how many distinct opportunities there are to leverage GPT technology:

1. Brainstorm analogies for explaining difficult concepts.
2. Generate a short-form content summary of takeaway learnings.
3. Expand the summary into a full draft of the chapter.
4. Rewrite the lesson as per standard lesson plan templates.
5. Write supporting notes for answering student questions on related concepts.
6. Find references to learn more related to the topic on the internet.
7. Draft a rubric for assessment across multiple learnt concepts with multiple competency levels.
8. Generate a bank of exam questions and model answers matching the rubric.
9. Present the content interactively using a chatbot teacher for self-paced learning.
10. Tidy up transcripts of recorded training videos and turn them into accessible training material for the hearing- and / or vision-impaired.

"You are terminated" or "I will be back"?

Technological civilization in its current form is moving inexorably towards a decisive moment, a crucial turning point when old things change and new things begin, a perilous situation?when things suddenly might start to go awry, a time when one should be especially wary, quick-witted and resourceful.

Many elements are at play in the dynamic of this situation's unfolding. While it might either turn out for better or for worse,?the latent possibility of a highly undesirable outcome?must remain in mind as we move forward.

Historically, such situations focus individuals as well as organizations sharply on what really matters. Incentives and motivations change, getting the collective adrenaline flowing,?potentially leading to new cooperative behaviors and even to the creation of new systems or structures.

Provided we have the courage to question and challenge traditional approaches and paradigms, the circumstances?act as a forcing function, leading to?emergence of new talent, rapid problem solving, cooperation among former rivals, dramatic policy shifts, sustainable systemic change, and increased resiliency for the next event.

Which way do you swing on this?