If Apple is "Behind" on AI, That is Just Apple Being Apple

Though Apple is widely considered to be “behind” in generative AI leadership, that perception is likely misplaced. Recall Apple’s traditional approach to technology innovation: it rarely is the “first” to deploy any new technology. Instead, it has excelled at packaging new technology in better, more user-friendly or elegant ways. 

In fact, it would have been a shock had Apple emerged early as a generative AI leader. 

Where Apple should emerge as a force is on-device AI, given its leadership in devices and device functions, where AI already has been deployed to support smartphone operations related to imaging and cameras; user voice input; voice-to-text translation or facial recognition. 

Use Case	Description
Facial Recognition (Unlocking Phones)	Faster and more secure authentication compared to server-based verification.
Image/Video Processing (Filters, Editing)	Real-time filters and effects applied directly on the device, without needing to upload and download media files.
Voice Recognition (Offline Assistants)	Offline access to voice commands for basic tasks like setting alarms or making calls.
Sensor Data Analysis (Fitness Trackers)	Real-time processing of biometric data for personalized health insights and fitness coaching.
AR/VR Applications (Overlays, Interactions)	Enhanced responsiveness and lower latency for a more immersive augmented or virtual reality experience.

On-Device AI Processing Advantage	Value
Faster Response Times	No need to send data back and forth to the cloud, leading to quicker results, especially for real-time applications.
Lower Power Consumption	Processing data locally reduces reliance on network connectivity, saving battery life on mobile devices.
Improved Privacy, Security	User data stays on the device, minimizing privacy concerns and potential security risks associated with cloud storage.
Lower Power Consumption	Less reliance on cloud servers translates to better battery life for personal devices.
Offline Functionality	Works even without an internet connection, essential for situations with limited access.
Offline Functionality	AI features can still function even without an internet connection, crucial for areas with unreliable network coverage.

AI is Gold Rush, but Maybe Not Wild West

Though both the internet and artificial intelligence are likely to wind up being categorized as general-purpose technologies, the early structure of the markets arguably is different. 

The early internet was a “Wild West” dominated by startups with unclear revenue models, facing low barriers to entry and few real gatekeepers. 

In contrast, the early AI market (especially generative AI) features more leadership by established firms with relatively clear ideas about possible revenue models, higher barriers to entry because of the costs of building models and training them, suggesting gatekeepers could emerge faster. 

Aspect	Early Internet	Early Generative AI Market
Leadership by startups or established Firms	Startups dominated innovation. Lower costs and agility gave them an edge.	Established firms with data and resources are playing a major role alongside AI startups.
Revenue Models	Unclear initially. Ads and e-commerce emerged.	Virtually every existing user experience, application and use case can be AI-enabled. Subscriptions, premium products and commerce and advertising all are conceivable..
Barriers to Entry	Relatively low, as virtual products had low production costs and low geographic barriers	Higher barriers. Requires access to large data troves, substantial computing power
Gatekeeper Power	Limited. Permissionless innovation was the model.	Tech giants such as Google, Amazon, Meta, Microsoft (and Apple, despite some current thinking) hold assets, with clear paths to monetize search, commerce, social media, enterprise and consumer software and devices.

And despite some early fears about monetization, digital infrastructure suppliers already are showing substantial revenue generation. App providers are rushing to add generative AI features to all existing products, with monetization based on potential to gain market share; create new premium products with AI features for additional costs; subscriptions; higher-value ad placement, operating cost and customer interaction advantages. 

All this suggests generative AI will find a faster and clearer road to monetized deployment. The AI market is less "Wild West" but still a "Gold Rush."

Governments Likely Won't be Very Good at AI Regulation

Artificial intelligence regulations are at an early stage, and some typical areas of enforcement, such as copyright or antitrust, will take some time to develop, in the former case because a sufficient body of precedent from prior cases must develop; in the latter case because markets will not be developed enough to make commercial power determinations. 

In the meantime, regulators seem to be focusing on the procedural areas: consumer safety (AI use for autonomous vehicles); privacy (when facial recognition can be used) ; security (use of personal data); transparency (how models have used training data); algorithmic bias; or liability (who is responsible if harm occurs?). But we remain early in those processes as well. 

Area of Interest	Description	Example Regulations
Safety and Security	Focuses on mitigating risks associated with AI systems, such as malfunctions, biases, and security vulnerabilities.	Mandates for robust testing and safety assessments of high-risk AI systems. - Regulations on the use of AI in critical infrastructure or autonomous vehicles.
Privacy	Protects individual privacy rights in the context of AI data collection, use, and decision-making.	Alignment with existing data privacy laws (e.g., GDPR, CCPA). - Restrictions on AI systems that use personal data for profiling or decision-making.
Transparency and Explainability	Aims to make AI systems more transparent and understandable, allowing for human oversight and accountability.	Requirements for developers to disclose how AI systems function and the data they use. - Right to explanation for individuals impacted by AI-driven decisions.
Algorithmic Bias	Addresses the potential for bias in AI algorithms due to training data or design choices.	Regulations on fairness and non-discrimination in AI algorithms used for hiring, loan approvals, etc. - Auditing of AI systems to identify and mitigate bias.
Liability and Accountability	Defines who is responsible for the actions and decisions of AI systems, particularly in cases of harm.	Clarification on liability for accidents or errors caused by autonomous AI systems. - Establishment of responsible actors in the AI development and deployment process.
Intellectual Property	Addresses ownership and rights related to AI creations (e.g., inventions, copyright in creative outputs).	Clarification on patentability of AI-generated inventions. - Determination of authorship rights for creative content produced by AI.

Some of us remain skeptical about government ability to usefully regulate important and profound new technologies, so hopefully overreach will not happen. 

Title	Authors	Publication	Key Takeaways
The Myth of the Digital Regulatory State	Sheila A. Brennan & Michael J. Meurer	Virginia Journal of Law & Technology	Argues that rapid technological change outpaces traditional regulatory frameworks. - Regulations often struggle to keep up with the evolving nature of new technologies.
Regulating Artificial Intelligence: A Multi-Stakeholder Approach	Daniel W. Drezner	Global Policy	Highlights the complexity of AI and the difficulty in defining and addressing potential risks. - Suggests a multi-stakeholder approach involving governments, industry, and civil society.
Can Technology Be Governed?	David Kaye & Debbie Sykes	Daedalus	Discusses the challenges of applying traditional governance structures to complex, globalized technologies. - Raises concerns about democratic control and potential for censorship.
The Innovation Paradox: How Global Governance Fails to Catch Up with Technological Change	Robert Falkner	Brookings Institution	Examines how rapid innovation creates challenges for international cooperation on regulation. - Suggests the need for more flexible and adaptable regulatory approaches.
The Case for Algorithmic Regulation	Daniel Kreutzfeldt	Oxford Internet Institute	Proposes a focus on algorithmic decision-making processes rather than specific technologies. - Argues for transparency and accountability in algorithms used by companies and governments.

Users Overestimate Value of IT

User perceptions often do not match underlying realities, whether that is the relative strengths of a sports team or the value of applying generative artificial intelligence. Overconfidence bias is one explanation. 

When surveyed or asked, people tend to overestimate their own abilities and knowledge, which can lead to an inflated perception of the impact of IT on their work. 

Confirmation bias also is at work. Confirmation bias essentially leads us to favor information that confirms our existing beliefs and downplay or ignore information that contradicts them. That obviously always is at work when decisions are made that have financial or strategic implications. Those who made the decisions have a vested interest in positive outcomes. 

The adage “You don’t get fired for recommending IBM” might be extended to any number of other buying decisions made related to IT. Risk aversion makes sense if your job might be imperiled by a bad decision. 

Going with firms with proven track records--rather than new upstarts--often makes perfect sense. The established “name” providers might cost more, or might not provide as much innovation, but the risk of unexpected failures is reduced. Backwards compatibility, domain knowledge, staff familiarity and training costs, conversion costs or ecosystem richness are other possible advantages of sticking with the known. 

The point is that users often overestimate the quantifiable advantages from new IT deployments and approaches, even as independent studies find mixed evidence of such clear improvements, at least within five to 10 years of a major technology deployment. 

Study Title	Authors	Year	Findings
"The Productivity Paradox in Information Technology"	Brynjolfsson, Erik and Lorin M.	1997	Found no clear evidence that IT adoption led to significant increases in overall worker productivity in the US economy.
"The Impact of Enterprise Resource Planning Systems on Firm Performance"	Brynjolfsson, Erik, Lorin M.	2002	Examined the impact of ERP systems on firms and found that while they can improve efficiency in some areas, the overall impact on profitability was mixed.
"From Hype to Reality: Exploring the Real Business Value of CRM Technology"	Reinartz, Werner and	2002	Studied the impact of Customer Relationship Management (CRM) systems and found that successful implementation required significant organizational changes beyond the technology itself.
"Does Investment in IT Really Pay Off?"	Brynjolfsson, Erik, Lorin M.	2016	Argued that the benefits of IT are often overstated, and that successful implementation requires a focus on complementary organizational changes and investments in worker skills.
"User Beliefs About Technology Usage: The Case of Enterprise Systems"	Dennis, Alan R. and Barbara	2001	Investigated user perceptions of enterprise systems and found a disconnect between user beliefs about the systems' capabilities and their actual effectiveness.

Also, most new IT implementations, especially those with important strategic impact, take time to bear fruit, as often, tangible outcomes require reshaping of whole business processes. So there often is a lag between the time new technologies are implemented and the time when people can cite clear positive outcomes. 

So watch for an avalanche of end user studies claiming benefits from deploying generative AI, for example, that cannot be independently verified, in terms of magnitude of gain. Vendors will want such attitudes and those making the spending decisions will want to keep their jobs. 

SKT Takes Arguably-Unusual AI Approach to Telco AI

SKT, which arguably is positioned differently from most telcos (it has business units that produce semiconductors) also is taking a for-now unusual approach to generative artificial intelligence, developing and supporting multiple large language models supporting telco operations as a vertical and in the Korean language initially. 

SKT has an in-house LLM called 'A.X' and also will support OpenAI's GPT and Anthropic's Claude, apparently aiming to use all three by supporting different use cases. 

It is possible one LLM might handle customer service; a second LLM might be used to support network operations and a third might handle back office operations, for example. 

Ignoring questions that inevitably will be raised about whether this makes ultimate sense, we already might not that each major player in the model business (as well as some expected entrants) already is slanting development in ways that support the existing core business models and revenue streams of each company. 

For example, Apple will undoubtedly focus on device AI, as many use cases will revolve around image processing, language translation or speech-to-text, plus ways to anticipate a user’s needs in context, for example. That makes sense as Apple is a device company, first and foremost. 

Meta is more likely to emphasize content curation, content moderation and natural language interfaces to content, given its roles in social media and user-generated content. 

Company	Core Business Model	AI Focus/Strategy
Apple	Devices and & Services (iPhone, Mac, App Store)	Device Intelligence: AI for on-device processing tasks like image recognition (Face ID), voice assistants (Siri), and personalized recommendations.  Software Optimization: AI for optimizing battery life, app performance, and user experience across Apple devices.
Meta (Facebook)	Social Networking and Advertising	Content Curation and Recommendation: AI for personalizing news feeds, suggesting friends and groups, and targeted advertising based on user data.  Natural Language Processing (NLP): AI for improving chatbots, sentiment analysis, and content moderation.
Microsoft	Cloud Computing (Azure),  Productivity Software (Office 365)	Enterprise AI Solutions: AI-powered tools for data analytics, automation, and productivity within businesses (Power BI, Dynamics 365).  Cognitive Services: Cloud-based AI services for developers to integrate features like speech recognition, computer vision, and translation into their applications.
Google	Search and  Advertising	Search and Ranking Algorithms: AI for refining search results, understanding user intent, and delivering highly relevant advertisements.  Voice Assistant (Google Assistant): AI for natural language interaction, voice-based device control, and smart home integration.
Amazon	E-commerce and  Cloud Computing (AWS)	Recommendation and Personalization Engines: AI for suggesting products, optimizing search results, and tailoring user experiences on Amazon.com.  Logistics and Supply Chain Optimization: AI for optimizing warehouse operations, predicting demand, and managing delivery routes.

Microsoft has long been a leader in enterprise software and productivity suites, so naturally will focus on enterprise AI and business solutions. 

Google will naturally reinforce its core search functions using LLM means to improve Google search abilities to answer questions and provide information more precisely, with greater context awareness and relevance, in more natural ways. That, in turn, can provide value to advertisers by improving the contextual placement of ads. 

Amazon has already moved to use LLMs for personalization and recommendations, product search functions in its e-commerce business, while moving to support LLM hosting as part of its Amazon Web Services business. 

More of this is going to happen as industry-specific and function-specific versions of LLMs develop. What remains to be seen is how many other relevant customizations could also develop.