How To Write About AI: A Beginner’s Guide

How do you write about AI — without mess­ing up?

Many PR pro­fes­sion­als and com­mu­nic­at­ors are tasked with writ­ing art­icles and texts about arti­fi­cial intel­li­gence.

Since I’ve writ­ten sev­er­al AI art­icles, I want to share some insights that could help you write your next piece on AI. ¹Please note that approx. 70% of this art­icle was writ­ten using premi­um GPT‑3 soft­ware.

Here we go:

Avoid Over-Inflating AI Concepts

Leading AI sci­ent­ists aren’t in agree­ment on how to describe arti­fi­cial intel­li­gence. As an art­icle writer, you could define AI in whatever way you see fit and be sure there are per­spect­ives to sup­port your chosen approach.

Do you want to describe a toast­er as an AI device? Sure, a toast­er is a machine that reacts to input to pro­duce a pre­dict­ive output.

But be care­ful. The ambi­gu­ity with­in the AI field is both a bless­ing and a curse for any­one writ­ing AI art­icles. You’re free to be cre­at­ive in your approach, but unfor­tu­nately, there’s noth­ing to hold onto.

The gen­er­al uncer­tainty means the PR writer must choose a con­sist­ent path through the AI text.

There’s an “arti­fi­cial intel­li­gence infla­tion” with­in cor­por­ate writ­ing since it’s easy to call just about any­thing “intel­li­gent” or “smart,” many tech brands over-use the AI term, espe­cially in their marketing.

My advice would be to adopt a more con­ser­vat­ive approach.
But what should that approach be, exactly?

If it’s an algorithm, call it an algorithm.
If it’s machine learn­ing, call it machine learn­ing.
If it’s a neur­al net­work, call it a neur­al network.

And so on.

What AI Are You Writing About?

Artificial intel­li­gence is an area of arti­fi­cial intel­li­gence research cov­er­ing the full poten­tial of future tech­no­lo­gies. This includes the vari­ous implic­a­tions of this tech­no­logy on soci­ety, employ­ment, and eco­nom­ic development. 

It also cov­ers how eco­nom­ics could play a role in how dif­fer­ent coun­tries may build the first AI to become an inde­pend­ent entity. Countries like Japan, China, and India have sig­ni­fic­antly advanced AI research and devel­op­ment through private enter­prises and gov­ern­ment participation.

There are attempts at clas­si­fy­ing dif­fer­ent types of AI. Since we can’t pin­point a gen­er­al defin­i­tion, defin­ing sub-types becomes challenging.

1. Reactive AI

A react­ive AI receives input and pro­duces a pre­dict­able out­put. IBM:s com­puter Deep Blue which defeated chess grand­mas­ter Gary Kasparov is a fam­ous example. ²Gary Kasparov is a chess legend. On September 10 1985, he faced off against the IBM machine, Deep Blue, to determ­ine wheth­er or not machines could out­per­form humans in chess. The world watched … Continue read­ing

Calling Deep Blue an AI won’t cause any raised eye­brows. It beat a human chess cham­pi­on, after all! However, what about the Netflix algorithm? What about a pock­et cal­cu­lat­or or a smart­phone? Or what about… a toaster? 

2. Limited Memory AI

Applications that use lim­ited memory AI are like react­ive AI sys­tems, but the dif­fer­ence is that a bit of memory AI will use his­tor­ic­al res­ults to self-correct.

If you’ve been tasked with writ­ing a text about AI, this is most likely the kind of AI you’ll write about. Almost all “AI applic­a­tions” that exist today fall under this category.

3. Theory of Mind AI

A the­ory of mind AI will under­stand you — at least to a degree. These tech­no­lo­gies are essen­tial to push AI into a space that could bene­fit humanity.

Is it pos­sible to build machines that can feel, both for them­selves and on behalf of oth­ers? It seems so, but it isn’t straightforward.

This is where the fam­ous Turing test comes into play; when will we be able to com­mu­nic­ate with a machine without telling it apart from how a human com­mu­nic­ates? ³The Turing test is a test of arti­fi­cial intel­li­gence (AI) developed by Alan Turing in 1950. The hypo­thet­ic­al test is con­duc­ted by an inter­rog­at­or who asks nat­ur­al lan­guage ques­tions to humans and AI … Continue read­ing

4. Self-Aware AI

Also known as the “sin­gu­lar­ity,” a machine could the­or­et­ic­ally become self-aware. 

The space of self-aware AI is highly hypo­thet­ic­al because we have yet to under­stand wheth­er humans are genu­inely con­scious (we could feel aware).

Three More Types of AI

Three more typ­ic­al clas­si­fic­a­tions can be help­ful for any­one about to write about AI.

5. ANI (Artificial Narrow Intelligence)

Artificial nar­row intel­li­gence, or ANI, is a term that applies to any com­puter that has been pro­grammed to do one thing very well. 

For example, the ANI chess pro­gram Deep Blue beat its human oppon­ents by ana­lys­ing more moves in a short­er time than human play­ers can. Many indus­tries use arti­fi­cial nar­row intel­li­gence for tasks ran­ging from facial recog­ni­tion to assem­bling automobiles.

ANI is a computer’s abil­ity to com­plete a nar­rowly-defined cog­nit­ive task. The term was coined by MIT sci­ent­ist and invent­or Marvin Minsky in 1966. ⁴Marvin Minsky was born in New York City on September 9, 1927. He atten­ded Harvard University, where he stud­ied math­em­at­ics and phys­ics. Later, while attend­ing Princeton University, he developed an … Continue read­ing

6. AGI (Artificial General Intelligence)

Artificial gen­er­al intel­li­gence (AGI) pro­duces a machine with an arti­fi­cially intel­li­gent sys­tem that can per­form tasks that require human intel­li­gence. One of the most chal­len­ging tasks is the rep­lic­a­tion of human reas­on­ing abilities. 

The cur­rent best-per­form­ing arti­fi­cial intel­li­gence is Deep Q‑network (DQN). A DQN agent will orches­trate its learn­ing pro­ced­ure and decide when to use rein­force­ment learn­ing. ⁵A Deep Q‑network, or DQN, is a rein­force­ment learn­ing algorithm that suc­cess­fully solves vari­ous engin­eer­ing prob­lems. The archi­tec­ture com­bines a super­vised learn­ing mod­el and an unsu­per­vised … Continue read­ing

7. ASI (Artificial Superintelligence)

Artificial Superintelligence is cre­at­ing a machine that has cog­nit­ive abil­it­ies that are super­hu­man to human beings. 

An AI sys­tem can be powered by a data­base that stores all the inform­a­tion, but it’s not enough for an AI to have this inform­a­tion. It also needs to think ration­ally and logic­ally, no mat­ter what. 

The fear is that AI will sur­pass human intel­li­gence and become too power­ful to control.

The Fear of AI

There are three types of fear when it comes to AI. Understanding these fears can be help­ful when writ­ing about arti­fi­cial intelligence.

Fear of Uncontrolled AI

An AI sys­tem could get out of con­trol. This goes for react­ive AI, lim­ited memory AI and ANI. Ironically, this is a fear of an AI that we can’t reas­on with — because it doesn’t under­stand. It just keeps doing what it’s been pro­grammed to do. 

Fear of AI Supremacy

If an AI becomes aware (the sin­gu­lar­ity) or thinks it is con­scious, it could start to reas­on dan­ger­ously for humans. We rarely con­sider the rights of insects, so why should an AI con­sider the rights of humans?

There are legit­im­ate con­cerns that arti­fi­cial intel­li­gence could be the next fron­ti­er of nat­ur­al evol­u­tion and that mech­an­ic­al life will sur­pass and even­tu­ally replace bio­lo­gic­al life.

Deterministic Fear

There’s a fear that con­scious­ness is an illu­sion, that we’re noth­ing more than bio­lo­gic­al machines. The cre­ation of seem­ingly self-aware machines poten­tially attacks our ideas of hav­ing free will.

Different AI Technologies

As you attempt to write about AI, you will face the chal­lenge of describ­ing how the tech­no­logy works. Below are a few terms that might be use­ful to research fur­ther before start­ing your writ­ing project.

Algorithms

Algorithms are used to make decisions. They are sequences of com­pu­ta­tion­al instruc­tions that are used to solve a problem. 

Algorithms are com­monly applied in com­puter sci­ence but are also used in areas out­side of com­puter sci­ence, such as genet­ics. They are used for everything from cal­cu­lat­ing the fast­est route to drive to work every morn­ing to match­ing donors with recip­i­ents for organ transplants.

Machine Learning

Machine learn­ing is a form of AI that allows com­puters to learn new inform­a­tion without being expli­citly programmed. 

A machine learn­ing sys­tem con­sists of two essen­tial parts: an algorithm and train­ing data. The algorithm finds pat­terns in the train­ing data and pre­dicts new data.

Neural Networks

Neural net­works are AI tech­no­lo­gies that mim­ic how the human brain works. They con­sist of a series of inter­con­nec­ted neur­ons that com­mu­nic­ate and pro­cess information. 

When presen­ted with a prob­lem, the net­work adjusts its con­nec­tions and thresholds until it solves it, comes up with a viable solu­tion, or runs out of time. 

Scientists can train neur­al net­works to per­form tasks by sup­ply­ing them with many examples and let­ting them fig­ure out the pat­terns independently.

Deep Learning

Deep learn­ing is a sub­set of machine learn­ing that is revolu­tion­ising the field of AI. Deep learn­ing ensures that machines learn in ways sim­il­ar to how humans do. 

Many dif­fer­ent types of algorithms can be used for deep learn­ing. Still, they all have one thing in com­mon: they auto­mat­ic­ally learn by pro­cessing large data sets and iter­at­ively improv­ing per­form­ance on some tasks.

Quantum Computing

Quantum com­puters are a new branch of com­put­ing that uses quantum bits, or qubits, instead of the tra­di­tion­al bin­ary bits. 

The primary advant­age of a quantum com­puter is its abil­ity to store expo­nen­tially more inform­a­tion than a con­ven­tion­al com­puter. A slight dif­fer­ence in tem­per­at­ure between two qubits can change their energy states which causes them to be either “up” or “down” at any giv­en time. This prop­erty is known as quantum entanglement.

AI in Pop Culture

What will hap­pen to humans when arti­fi­cial intel­li­gence can do everything bet­ter than becom­ing more and more pre­val­ent in the 21^st century? 

Many have seen arti­fi­cial intel­li­gence as a threat to human civil­isa­tion, with some call­ing for it to be banned. But oth­ers argue that an AI-driv­en world could be our only hope.

Artificial Intelligence has been a pop­u­lar top­ic in pop cul­ture for quite some time now. Movies have depic­ted AI as either friendly or evil. Still, it’s not until recently that soci­ety has begun to under­stand what it means for human­ity if the sin­gu­lar­ity is created.

The gen­er­al myth­os sur­round­ing AI can be used for ref­er­ences to make your AI text more relat­able or understandable.

Skynet

In the Terminator movies, the Skynet AI doesn’t show self-aware­ness signs. It behaves more like a lim­ited memory AI applic­a­tion for war games that run amok and thinks of humans as enemies. 

The Terminator T‑800, spawned by Skynet and hijacked by human res­ist­ance, is sent back in time and slowly becomes self-aware through­out the franchise.

Mr Smith

The AI in The Matrix is fully self-aware. While it struggles to under­stand how to reg­u­late human­ity, its many pro­grams can con­nect emo­tion­ally with the human characters.

Mr Smith is such a sep­ar­ate pro­gram tasked with man­aging the com­puter-gen­er­ated world. Ironically, he learns he hates his job almost as much as humans.

HAL 9000

Like the T‑800, Hal 9000 ranks in between Skynet and Mr Smith. 

While the T‑800 and Mr Smith are self-aware, they can reas­on out­side their hard-coded para­met­ers. HAL 9000, on the oth­er hand, would eas­ily pass the Turing test but seems unable to break free from its pro­gram­ming. Instead, HAL 9000 finds cre­at­ive ways to inter­pret its mis­sion parameters.

Summary: How To Write About AI

Use a humble ton­al­ity and reas­on cau­tiously. Even the most prom­in­ent sci­ent­ist in AI isn’t in agree­ment. Be mind­ful of present­ing any cer­tain­ties. Be extra humble if you’re writ­ing about react­ive- or lim­ited memory AI.

Know your primary AI clas­si­fic­a­tions. You’ll most likely write about react­ive- or lim­ited memory AI. Don’t lead the read­er to think that the tech­no­logy you cov­er is first-cous­in with the singularity.

Take your time explain­ing the tech­no­logy. Any read­er of your text about AI will want to know how the tech­no­logy works. This might be the most chal­len­ging part for the writer, but it’s cru­cial for mak­ing your text work.

Some of your read­ers fear AI. The fear of AI is real; even highly edu­cated experts and intel­lec­tu­als express their con­cerns. Use fear to make your text more com­pel­ling, but do so spar­ingly — espe­cially if you’re writ­ing about ANI (arti­fi­cial nar­row intelligence).

AI mat­ters to all of us. AI has been a part of our pop cul­ture for a long time. Use ref­er­ences to make your text stand out, but remem­ber that sci­ence fic­tion is still… science-fiction.

PR Resource: Free Writing PR Course