New Scientist - USA (2022-06-04)

14 | New Scientist | 4 June 2022

News

“A combination of a coated

drug and radiotherapy

eliminated tumours

in 67 per cent of mice”

Health

COATING a cancer drug with

nanoparticles can help it cross the

barrier that protects the brain from

chemicals elsewhere in the body,

enabling it to unleash an immune

response that kills brain tumour

cells. In a small study in mice, the

encapsulated drug – used with

radiotherapy – eliminated tumours

in two-thirds of the animals.

Glioblastoma is an aggressive

type of tumour that forms when

brain cells called glial cells grow

out of control. These tumours

are usually treated with surgery,

followed by radiotherapy and

sometimes chemotherapy.

However, while some drugs have

shown promise at boosting the

body’s immune response against

brain tumours, the drugs often

struggle to cross the blood-brain

barrier to reach the tumour cells.

Maria Castro at the University

of Michigan Medical School and

her colleagues have now tested

whether encapsulating a drug in

synthetic nanoparticles – made with

proteins found in the body – could

help it reach the brains of mice.

The team injected mice

with either a drug alone or a

nanoparticle-coated version. Eight

hours later, mice that received the

coated treatment had three times

more of the drug in their brains

than mice given a normal dose.

Castro’s team then tested how

well the drug – with and without

the nanoparticle coating – worked

in mice that had tumour cells

implanted into their brains.

A combination of the

nanoparticle-coated drug and

radiotherapy eliminated tumours

in 67 per cent of mice, enabling

them to survive for 90 days, on

average, from the moment tumour

cells were implanted. Mice that

received either radiotherapy or

the encapsulated drug alone

died after about 60 days (ACS

Nanotechnology, doi.org/gp7ndb).

The treatment still has to be tested

to see if it works in other animal

models of glioblastoma, before

it could be tested in humans.  ❚

Immunotherapy for

brain cancer can be

made more effective

Carissa Wong

Analysis Artificial intelligence

Will text-to-speech AIs put illustrators out of a job?

Google’s Imagen AI produces high-quality images from a text

prompt, but is it ready to replace humans, asks Alex Wilkins

TECH firms are racing to create

artificial intelligence algorithms

that can produce high-quality

images from text prompts, with the

technology seeming to advance so

quickly that some predict that human

illustrators and stock photographers

will soon be out of a job. In reality,

limitations with these systems mean

it will probably be a while before they

are available to the general public.

Text-to-image generators that use

neural networks have made great

progress in recent years. The latest,

Imagen from Google, comes hot on

the heels of DALL-E 2, which was

announced by OpenAI in April.

Both models use a neural

network trained on a large number of

examples to categorise how images

relate to text descriptions. When

given a new description, the network

repeatedly generates images, altering

them until they most closely match

the text based on what it has learned.

While the images presented by

both firms are impressive, researchers

have questioned whether the results

are being cherry-picked to show

the systems in the most favourable

light, a common practice. “You need

to present your best results,” says

Hossein Malekmohamadi at

De Montfort University in the UK.

One problem in judging these

AI creations is that both firms have

declined to release public demos

that would allow researchers and

others to put them through their

paces. Part of the reason for this is

a fear that the AI could be used to

create misleading images, or simply

that it could generate harmful results.

The models rely on data sets

scraped from large, unmoderated

portions of the internet, such as

the LAION-400M data set, which

Google has said is known to contain

“pornographic imagery, racist slurs,

and harmful social stereotypes”.

The researchers behind Imagen

say that as they can’t guarantee it

won’t inherit problematic content,

they can’t release it to the public.

OpenAI claims to be improving

DALL-E 2’s “safety system” by

“refining the text filters and tuning

the automated detection & response

system for content policy violations”,

while Google is seeking to address

the challenges by developing a

“vocabulary of potential harms”.

Neither firm was able to speak to

New Scientist before publication.

Unless these problems can be

solved, it seems unlikely that big

research teams like Google or

OpenAI will offer their text-to-image

systems for general use. It is possible

that smaller teams could choose to

release similar technology, but the

sheer amount of computing power

required to train these models on

huge data sets tends to limit work

on them to big players.

Despite this, the friendly

competition between the big firms

is likely to mean the technology

continues to advance rapidly,

as tools developed by one group

can be incorporated into another’s

future model. For example, diffusion

models, where neural networks learn

how to reverse the process of adding

random pixels to an image in an

effort to improve them, have shown

promise in machine-learning models

in the past year. Both DALL-E 2 and

Imagen rely on the technique, after

it proved effective in less-powerful

models, such as OpenAI’s Glide

image generator. ❚

IMA

GE

N/G

OO

GL

E

Six examples of images

generated by Google’s

Imagen AI