Best Uses for Artificial Intelligence and Machine Learning

Communication and email

Email is still the dominant business communication channel, and AI now writes a lot of it. Smart Reply in Gmail and Outlook suggests three quick responses based on the message content; Smart Compose auto-completes phrases as you type. The 2023-2025 wave added LLM-powered drafting in Google Workspace (Gemini), Microsoft 365 (Copilot), and Apple Mail (Apple Intelligence), which can draft full replies, summarize long threads, and adjust tone on demand. Note: Google’s Allo messaging app was discontinued in March 2019; the AI-reply features moved into Gmail and Google Messages.

Mobile banking deposits

Mobile check deposit is one of the most-used everyday AI features. Most U.S. banks (Chase, Bank of America, Wells Fargo, Capital One, and others) use computer-vision and ML systems built on Mitek’s technology to recognize handwriting, read MICR routing numbers, validate signatures, and process check deposits from phone photos. Same underlying tech now powers digital wallet onboarding (driver’s-license scans), insurance claims (damage photos), and ID verification across financial services.

Academic honesty and AI-detection

Plagiarism detection has been computational for years (Turnitin, Copyleaks, Plagscan). The bigger 2024-2026 question is AI-detection: identifying student work generated by ChatGPT, Claude, Gemini, or other LLMs. Tools like GPTZero, Originality.ai, and Turnitin’s AI-detection module attempt this with imperfect accuracy — false positives are common, especially for non-native English writers, and detection accuracy degrades as LLMs improve. Most universities have shifted from outright bans toward policies on disclosure, AI-aware assignment design, and proctored in-class assessments.

Credit scoring and lending

Banks and lenders increasingly augment traditional credit scores with ML risk models that incorporate transaction patterns, employment history, and alternative data. Companies like Upstart, Zest AI, and most major banks’ in-house models claim better risk discrimination than FICO alone. The flip side: ML credit decisioning has drawn regulatory scrutiny — the CFPB, EEOC, and state regulators have issued guidance on disparate-impact testing, and the EU AI Act classifies credit-scoring as a high-risk AI system requiring conformity assessment, transparency, and human oversight.

Facebook (Meta) personalization and content moderation

Meta’s ML stack drives news-feed ranking, ad targeting, content recommendations on Reels, group suggestions, and content moderation. Facial recognition for photo tagging was wound down in late 2021 across Facebook (Meta retired the system, deleted associated face templates, and now uses other signals for tagging). The big 2024-2026 shift: Meta released the Llama family of open-weights LLMs and integrated Meta AI across Facebook, Instagram, WhatsApp, and Messenger as a chat assistant.

Shopping and product discovery

Recommender systems are the original mass-market ML success story. Amazon, Wayfair, Etsy, eBay, and the major retail platforms all use collaborative filtering plus deep-learning models to surface relevant products, personalize search, and assemble “customers also bought” sections. The 2024-2026 layer adds conversational shopping (Amazon’s Rufus, Walmart’s AI assistant) and AI image search (Google Lens, Pinterest Lens) where users can shop from a photo.

Chatbots and conversational AI

This is the most-transformed category since the original article was written. Pre-2022 chatbots used rule-based or narrow ML systems with limited language understanding. Post-ChatGPT (November 2022), large language models — ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Copilot (Microsoft), Llama (Meta) — handle open-ended conversation, code generation, summarization, translation, and creative writing at near-human quality. Customer-support chatbots built on these models (via Intercom Fin, Zendesk AI, Salesforce Einstein, and a flood of vertical SaaS layers) routinely resolve 30-60% of inbound tickets without human escalation.

Personalized shopping experiences

Beyond recommenders, ML powers individualized merchandising — different shoppers see different homepage hero modules, different category sequencing, different size and fit recommendations, even different pricing on some platforms. Companies like Bloomreach, Algolia, and Constructor.io let mid-market retailers add this kind of personalization without building it from scratch. Privacy regulations (GDPR, CCPA, the EU AI Act) place real limits on how aggressively this can be done in regulated jurisdictions.

Card-specific fraud prevention

Real-time fraud detection runs almost every credit and debit transaction through ML scoring before approval — Visa Advanced Authorization, Mastercard’s Decision Intelligence, and bank-side risk engines all use neural networks to score transactions on hundreds of features (geography, merchant category, time, device fingerprint, behavioral patterns) in milliseconds. Square, Stripe, and other modern payment processors layer their own ML on top. The system also reduces false declines (legitimate transactions wrongly blocked), which used to be a major source of merchant lost revenue.

Snapchat and image filters

Snapchat’s real-time facial filters (Lenses) use ML for face landmark detection, head-pose tracking, and image segmentation. Instagram, TikTok, and BeReal have similar capabilities; Apple’s Animoji and Memoji use the same family of techniques. The newer wave: generative AR filters that combine text prompts with face tracking (powered by diffusion models) and on-device LLMs that personalize filter behavior to a user.

Education and grading

Automated essay scoring is well-established for standardized testing — ETS’s e-rater, Pearson’s Intelligent Essay Assessor, and similar systems score alongside human graders for SAT, GRE, TOEFL, and state-level exams. The 2024-2026 shift: LLM-based AI tutoring (Khan Academy’s Khanmigo, Carnegie Learning, Google Classroom’s AI features), AI-assisted grading for short-answer and essay work, and platform-level integrity tooling that flags AI-generated submissions and scaffolds AI-aware assessment design.

Fraud detection beyond cards

Beyond payment cards: account-takeover detection, fake-account creation prevention, claim fraud in insurance, identity verification at signup, money-laundering pattern detection in compliance. FICO Falcon, SAS Fraud Management, Feedzai, and bank-internal systems run anomaly detection plus deep-learning models that surface suspicious behavior in near real time.

Spam filtering

Inbox spam filtering is one of the longest-running production ML applications. Modern Gmail, Outlook, and iCloud Mail use deep-learning classifiers that incorporate sender reputation, content features, embedded-link analysis, and user feedback to keep spam to a low single-digit percentage of mail. The same techniques now defend against phishing and increasingly against AI-generated spam (which sounds more like real human writing and is harder to flag on language alone).

Financial advice and robo-advisors

Robo-advisors automate investment portfolio construction, rebalancing, and tax-loss harvesting using ML. Betterment, Wealthfront, Schwab Intelligent Portfolios, and Vanguard Personal Advisor Services manage hundreds of billions of dollars combined. The 2024-2026 layer adds LLM-driven conversational financial planning — answering questions about retirement projections, tax implications, and goal trade-offs — while still routing actual securities decisions through licensed advisor frameworks where regulation requires it.

Pinterest and computer vision

Pinterest’s computer-vision system identifies objects, styles, colors, and contexts in images for search, recommendation, and shoppable-pin matching. The same family of techniques powers Google Lens, image search across e-commerce, content moderation, and accessibility (auto-generated alt text). Diffusion-model image generation (Midjourney, Stable Diffusion, DALL-E, Adobe Firefly) is reshaping the upstream side: a meaningful share of new images on visual platforms are now AI-generated.

Online recommendations

Beyond shopping: Netflix, Spotify, YouTube, TikTok, and the streaming platforms all run sophisticated recommender systems that decide what plays next, what gets surfaced on your homepage, and what notifications to send. TikTok’s “For You” algorithm is widely studied for the speed at which it adapts to individual user behavior. The recommender problem is now also a regulatory issue — the EU Digital Services Act requires platforms to offer users a non-personalized feed option.

Navigation and shorter commute times

Waze, Google Maps, and Apple Maps all use ML to predict traffic, suggest alternate routes, and incorporate live road conditions (accidents, construction, weather). Real-time fleet data, anonymized GPS pings from millions of phones, and historical patterns feed into neural-network models that estimate arrival times within meaningful accuracy. Public-transit apps (Citymapper, Transit) layer multimodal route optimization on top.

Voice and voice-to-text

“Hey Siri,” “OK Google,” “Alexa,” and Microsoft Copilot voice all use deep-learning speech recognition that achieves human-parity word-error rates for many accents and conditions. Whisper (OpenAI), Google Cloud Speech-to-Text, Apple’s on-device speech recognition, and competitors like AssemblyAI and Deepgram handle transcription at scale. Voice cloning and synthesis (ElevenLabs, OpenAI TTS) crossed into convincing-quality territory in 2023-2024 — useful for accessibility and audio production, troubling for impersonation fraud.

Aviation autopilots and assisted control

Modern aircraft autopilots have used computer-controlled systems for decades, with continued improvement in fuel efficiency, weather handling, and instrument readings. That said, the “pilots only fly for a few minutes per flight” framing oversimplifies — Boeing’s 737 MAX MCAS issues (2018-2019 grounding) and the January 2024 Alaska Airlines door-plug incident illustrate that automation enhancements augment rather than replace pilot judgment. Single-pilot and crewless commercial cargo flights are an active area of research and regulatory debate but not yet production reality.

Smart homes and connected devices

Smart homes are mature now, not aspirational. Apple HomeKit, Google Home, Amazon Alexa, and Samsung SmartThings let users control lighting, climate, locks, cameras, appliances, and entertainment from voice or app. The Matter protocol (launched October 2022 by the Connectivity Standards Alliance) finally gave the category a cross-vendor interoperability standard, which had been the long-running adoption blocker. ML increasingly powers automation routines (e.g., learning typical schedules and pre-conditioning rooms).

Ridesharing and dynamic pricing

Uber, Lyft, DoorDash, Grubhub, Instacart, and the gig-economy platforms all use ML for matching, pricing, and ETAs. Surge / dynamic pricing is computed by short-horizon supply-and-demand models. Routing optimizes for combined factors — driver acceptance probability, expected fare, road conditions, and aggregate platform-level objectives. Growing regulatory scrutiny in the EU and several U.S. cities targets pay floors, deactivation due process, and algorithmic transparency.

Personalized learning at scale

Adaptive-learning platforms (Khan Academy with Khanmigo, Duolingo, Carnegie Learning, DreamBox, Squirrel AI) use ML to adjust difficulty, pacing, and content selection per student. Post-ChatGPT, the category absorbed LLM-based tutoring that handles open-ended student questions, simulates Socratic dialogue, and generates personalized practice problems. Khan Academy’s Khanmigo (built on GPT-4) is the highest-profile public example.

Email categorization

Gmail’s automatic sorting (Primary, Promotions, Social, Updates, Forums) and similar features in Outlook and Apple Mail use ML classifiers trained on billions of emails to file incoming mail. Newer features auto-summarize threads (Gmail’s “Summarize this email”), suggest unsubscribes for low-engagement senders, and surface time-sensitive items. The same systems power calendar inference (auto-adding events from email content) when users opt in.

Smart assistants

The 2026 smart-assistant landscape: Siri (Apple, integrated with Apple Intelligence and ChatGPT for complex queries), Google Assistant / Gemini on Android (Google migrated Assistant features into Gemini in 2024), Alexa on Amazon devices and the AI-augmented Alexa+ (announced 2024), and Microsoft Copilot on Windows. Cortana was retired across most consumer products. Note: Alexa is Amazon’s assistant — it works on Android phones via the Alexa app, but the native Android assistant has always been Google’s (now Gemini).

Instagram and social-media ML

Instagram’s ML stack drives Reels recommendations, hashtag clustering, content moderation, accessibility (auto-generated alt text), and ad targeting. Generative-AI features added 2024-2025 include AI-assisted Reel editing, image generation in DMs, and AI characters / personas. Content moderation now also has to detect AI-generated content (deepfakes, synthetic profile photos) — a moving target as generation models improve.

Generative AI for writing and creative work

The biggest category that didn’t exist when this article was originally written. ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Microsoft Copilot, Llama (Meta), and Notion AI handle drafting, editing, summarizing, brainstorming, code generation, and translation across most professional contexts. The underlying technology is the large language model (LLM) — a transformer-architecture neural network trained on enormous text corpora to predict next tokens, then fine-tuned with reinforcement learning from human feedback (RLHF) for instruction-following and safety.

Image generation went mainstream in 2022-2023 with DALL-E 2/3, Midjourney, Stable Diffusion, and Adobe Firefly — diffusion models that turn text prompts into photoreal or illustrated images. Video generation followed in 2024-2025 with OpenAI Sora, Google Veo, Runway Gen-3, and Kling. Coding assistants (GitHub Copilot, Cursor, Codeium, Tabnine) are now in mainstream developer use, with several measurable productivity gains in published studies. The category has also reshaped knowledge work — research synthesis, contract drafting, customer support, and creative brainstorming now routinely involve LLM assistance as a first-draft layer.

Autonomous and assisted driving

Production self-driving services run in limited geographies — Waymo operates commercial robotaxi service in Phoenix, San Francisco, Los Angeles, and Austin (continuing geographic expansion 2024-2026, with international expansion announced 2025). GM’s Cruise paused operations after a 2023 incident with a pedestrian and was wound down in 2024. Tesla’s Full Self-Driving (Supervised) still requires driver supervision and remains the subject of ongoing NHTSA investigations; Tesla launched an unsupervised robotaxi pilot in Austin in 2025 with limited operational design domain. Chinese autonomous services from Baidu Apollo Go, Pony.ai, and WeRide operate in several Chinese cities at varying levels of safety-driver oversight.

Lower-stakes assisted-driving features — adaptive cruise control, lane-keeping assist, automatic emergency braking, blind-spot monitoring, automatic parking — are now standard or near-standard in most new vehicles and use ML for perception (camera and radar fusion) and decisioning. The IIHS and NHTSA have both pushed for AEB to become mandatory on new light vehicles, with regulatory phase-ins through 2029. The technical foundation across all of these systems is a combination of computer vision, sensor fusion, and ML-based behavior prediction — the same building blocks across vendors, with the differentiator being scale of training data and operational design domain.

Healthcare and medical imaging

ML for medical imaging is one of the most validated AI/ML applications — over 1,000 FDA-cleared AI/ML medical devices have been listed since the FDA began publishing the catalog (2024 list). FDA-cleared algorithms now assist radiologists with breast cancer detection in mammography (DeepHealth, iCAD, Lunit), retinal-disease screening (Google’s Verily, IDx-DR for diabetic retinopathy), pulmonary nodule detection in CT scans, stroke triage (Viz.ai), and dozens of other narrow tasks. Other production uses: drug-discovery screening (DeepMind’s AlphaFold released structures for nearly every known protein in 2022 — over 200 million predicted structures), wearable-based arrhythmia detection (Apple Watch ECG, Fitbit), clinical decision support (Epic’s sepsis prediction, though the latter has drawn validation criticism), AI scribes for clinical documentation (Abridge, Nuance DAX, Suki), and patient-triage chatbots that screen symptoms before routing.

AI safety, regulation, and ethics

The largest 2024-2026 shift in how organizations approach AI is the regulatory layer. The EU AI Act entered into force August 1, 2024 with phased compliance dates through 2026-2027; it classifies AI systems by risk (banned, high-risk, limited, minimal) and imposes conformity assessments, transparency requirements, and human-oversight obligations on high-risk systems including credit decisioning, employment screening, biometric identification, and law-enforcement use. Penalties for non-compliance scale up to €35 million or 7% of global annual turnover (whichever is higher) — meaningfully higher than GDPR.

The Biden Executive Order on AI (October 2023) was rescinded January 2025; the Trump administration issued new AI executive orders in 2025 with different priorities, including faster federal AI adoption and reduced regulatory burden on US-based developers. State-level AI laws have proliferated — Colorado’s SB24-205 (2024) and California’s SB-1047 (vetoed but successor bills under consideration) are the most-cited examples. Outside North America and Europe: the UK AI Safety Institute coordinates voluntary frontier-model testing; China’s Generative AI Measures (effective August 2023) regulate generative AI services; many other jurisdictions are at various stages of AI policy development. Most large organizations now run an AI governance program with documented data-handling, model-evaluation, human-in-the-loop controls, and risk classification mapped to the regulations they’re subject to.

Frequently asked questions

What’s the difference between AI and ML?

AI is the broader field — machines doing tasks that traditionally required human intelligence. ML is one approach within AI: algorithms that improve from data without being explicitly programmed for every case. Most production “AI” today is actually ML; some products combine ML with rule-based logic.

How has ChatGPT changed the AI landscape?

The November 2022 ChatGPT launch took LLM capabilities mainstream. Pre-2022 conversational AI was narrow and brittle; post-2022 LLMs handle open-ended conversation, code generation, summarization, and translation at near-human quality. Most consumer AI features added since 2023 are LLM-powered.

Is generative AI safe to use for sensitive content?

Depends on the tool tier. Free consumer tiers (default ChatGPT, default Claude) often retain prompts for training unless you opt out. Enterprise tiers (ChatGPT Enterprise, Claude for Work, Microsoft Copilot 365, Notion Enterprise, Google Workspace AI) typically include data-protection and no-training contractual commitments. For regulated industries or proprietary work, use the enterprise tier.

Does the EU AI Act apply to me?

If your AI system is placed on the EU market or its output is used in the EU, yes — even if your company isn’t headquartered there. The Act’s scope is broad, with risk classification driving the actual obligations. Credit scoring, employment screening, law-enforcement biometrics, and educational testing are flagged as high-risk and trigger the heaviest compliance burden.

What are the biggest AI/ML categories I should watch in 2026-2027?

Multimodal models (text+image+audio+video in one model), AI agents that can take multi-step actions, on-device AI (Apple Intelligence, Pixel AI features), AI-native search experiences (Perplexity, Google AI Overviews, ChatGPT search), and AI safety / governance tooling for the EU AI Act compliance wave.

The bottom line

AI and machine learning are no longer futuristic — they’re infrastructure. Most products people interact with daily, from email to navigation to social media to banking, contain ML. The 2022-2026 wave of generative AI added a new layer on top: language, image, audio, and video generation that’s good enough to ship in production tools. The two questions worth asking in 2026 are where AI is delivering value (the categories above are the well-validated ones) and under what governance (especially for regulated, high-risk, or user-facing applications subject to the EU AI Act and similar frameworks).