How does artificial intelligence learn? - Briana Brownell

Translator: Anna B

Today, artificial intelligence helps doctors diagnose patients,

I dag hjælper kunstig intelligens læger med at diagnosticere patienter,

pilots fly commercial aircraft, and city planners predict traffic.

piloter med at flyve kommercielle flyere og byplanlæggere med at forudsige trafik.

But no matter what these AIs are doing, the computer scientists who designed them

Men uanset hvad AI laver, så ved de dataloger, der har designet dem

likely don’t know exactly how they’re doing it.

sandsynligvis ikke præcist, hvordan de gør det.

This is because artificial intelligence is often self-taught,

Det er fordi, kunstig intelligens ofte er selvlært

working off a simple set of instructions

og arbejder ud fra et sæt af simple instruktioner

to create a unique array of rules and strategies.

for at skabe en unik række regler og strategier.

So how exactly does a machine learn?

Men hvordan lærer en maskine?

There are many different ways to build self-teaching programs.

Der er mange måder at bygge selvlærende programmer.

But they all rely on the three basic types of machine learning:

Men de er alle afhængige af de tre grundlæggende typer af maskinlæring:

unsupervised learning, supervised learning, and reinforcement learning.

Uovervåget læring, overvåget læring og forstærkende læring.

To see these in action,

For at se dem i aktion

let’s imagine researchers are trying to pull information

lad os forestille os, at nogle forskere forsøger at indhente information

from a set of medical data containing thousands of patient profiles.

fra et medicinsk datasæt, der indeholder tusindvis af patientprofiler.

First up, unsupervised learning.

Først har vi uovervåget læring.

This approach would be ideal for analyzing all the profiles

Denne tilgang ville være ideel til at analysere alle profilerne,

to find general similarities and useful patterns.

for at finde generelle ligheder og nyttige mønstre.

Maybe certain patients have similar disease presentations,

Måske har nogle patienter lignende sygdomspræsentationer,

or perhaps a treatment produces specific sets of side effects.

eller måske giver en behandling et specifikt sæt af bivirkninger.

This broad pattern-seeking approach can be used to identify similarities

Denne brede mønstersøgende tilgang kan bruges til at identificere ligheder

between patient profiles and find emerging patterns,

mellem patientprofiler og finde nye mønstre -

all without human guidance.

alt sammen uden menneskelig vejledning.

But let's imagine doctors are looking for something more specific.

Men forestil dig, at lægerne leder efter noget mere specifikt.

These physicians want to create an algorithm

Lægerne vil gerne lave en algoritme

for diagnosing a particular condition.

til at diagnosticere en bestemt sygdom.

They begin by collecting two sets of data—

De begynder med at indsamle to datasæt -

medical images and test results from both healthy patients

medicinske billeder og testresultater fra både raske patienter

and those diagnosed with the condition.

og dem, der er diagnosticeret med sygdommen.

Then, they input this data into a program

Derefter indtastes denne data i et program

designed to identify features shared by the sick patients

designet til at identificere træk fælles for de syge patienter,

but not the healthy patients.

men ikke hos de raske parienter.

Based on how frequently it sees certain features,

Baseret på hvor ofte den ser visse fællestræk,

the program will assign values to those features’ diagnostic significance,

vil programmet tildele værdier til trækkenes diagnostiske betydning,

generating an algorithm for diagnosing future patients.

og generere en algoritme til diagnosticering af fremtidige patienter.

However, unlike unsupervised learning,

Men i modsætning til uovervåget læring

doctors and computer scientists have an active role in what happens next.

har læger og dataloger en aktiv rolle i, hvad der sker herefter.

Doctors will make the final diagnosis

Lægerne stiller den endelige diagnose

and check the accuracy of the algorithm’s prediction.

og kontrollere nøjagtigheden i algoritmens forudsigelse.

Then computer scientists can use the updated datasets

Så kan datalogerne bruge de opdaterede datasæt

to adjust the program’s parameters and improve its accuracy.

til at justere programmets parametre og øge dets nøjagtighed.

This hands-on approach is called supervised learning.

Denne tilgang kaldes for overvåget læring.

Now, let’s say these doctors want to design another algorithm

Lad os nu sige, at lægerne vil at designe en anden algoritme

to recommend treatment plans.

til at anbefale behandlingsplaner.

Since these plans will be implemented in stages,

Siden planerne vil blive implementeret i etaper,

and they may change depending on each individual's response to treatments,

og de kan ændres afhængigt af den enkeltes reaktion på behandlingen,

the doctors decide to use reinforcement learning.

beslutter lægerne at bruge forstærkende læring.

This program uses an iterative approach to gather feedback

Dette program bruger en iterativ tilgang til at indsamle feedback

about which medications, dosages and treatments are most effective.

om hvilke lægemidler, doseringer og behandlinger, der er mest effektive.

Then, it compares that data against each patient’s profile

Derefter sammenlignes denne data med patientens profil,

to create their unique, optimal treatment plan.

for at give dem en unik og optimal behandlingsplan.

Efterhånden som behandlingerne skrider frem

As the treatments progress and the program receives more feedback,

og programmet modtager mere feedback,

it can constantly update the plan for each patient.

kan det konstant opdatere planen for hver patient.

None of these three techniques are inherently smarter than any other.

Ingen af disse tre teknikker er bedre end de andre.

While some require more or less human intervention,

Mens nogle kræver mere eller mindre menneskelig indgriben,

they all have their own strengths and weaknesses

har de deres forskellige styrker og svagheder,

which makes them best suited for certain tasks.

som gør dem bedst egnet til bestemte opgaver.

However, by using them together,

Men ved at bruge dem sammen

researchers can build complex AI systems,

kan forskere bygge komplekse AI-systemer,

where individual programs can supervise and teach each other.

hvor individuelle programmer kan overvåge og undervise hinanden.

For example, when our unsupervised learning program

For eksempel, når vores uovervågede læringsprogram

finds groups of patients that are similar,

finder grupper af patienter, der ligner hinanden,

it could send that data to a connected supervised learning program.

kunne det sende denne data til et tilsluttet overvåget læringsprogram.

That program could then incorporate this information into its predictions.

Dette program kunne så inkorporere denne information i sine forudsigelser.

Or perhaps dozens of reinforcement learning programs

Eller måske kunne snesevis af forstærkende læringsprogrammer

might simulate potential patient outcomes

simulere potentielle patientresultater

to collect feedback about different treatment plans.

for at indsamle feedback om forskellige behandlingsplaner.

There are numerous ways to create these machine-learning systems,

Der er mange måder at lave maskinlæringssystemer på,

and perhaps the most promising models

og de måske mest lovende modeller

are those that mimic the relationship between neurons in the brain.

er dem, der efterligner forholdet mellem neuroner i hjernen.

These artificial neural networks can use millions of connections

Disse kunstige neurale netværk kan bruge millioner af forbindelser på

to tackle difficult tasks like image recognition, speech recognition,

at tackle svære opgaver, som billedgenkendelse, talegenkendelse

and even language translation.

og endda sprogoversættelse.

However, the more self-directed these models become,

Men jo mere selvstyrende modellerne bliver,

the harder it is for computer scientists

jo sværere er det for dataloger

to determine how these self-taught algorithms arrive at their solution.

at bestemme hvordan de selvlærte algoritmer, når frem til deres resultater.

Researchers are already looking at ways to make machine learning more transparent.

Forskere ser allerede på måder at gøre maskinlæring mere transparent.

But as AI becomes more involved in our everyday lives,

Men efterhånden som AI fylder mere i vores hverdag

these enigmatic decisions have increasingly large impacts

har disse gådefulde beslutninger stadig større indvirkning

on our work, health, and safety.

på vores arbejde, sundhed og sikkerhed.

So as machines continue learning to investigate, negotiate and communicate,

Efterhånden som maskiner fortsat lærer at undersøge, forhandle og kommunikere,

we must also consider how to teach them to teach each other to operate ethically.

må vi også overveje, hvordan vi skal undervise dem i

at lære hinanden, hvordan man handler etisk.